[
  {
    "id": "anthropic-sitemap:_pdfs:096d94c1a91c6480806d8f24b2344c7e2a4bc666",
    "type": "document",
    "title": "096d94c1a91c6480806d8f24b2344c7e2a4bc666",
    "abstract": "--- source_url: https://www-cdn.anthropic.com/096d94c1a91c6480806d8f24b2344c7e2a4bc666.pdf referrer: https://www.anthropic.com/research/anthropic-economic-index-january-2026-report pages: 55 fetched_at: 2026-05-30T00:34:56.046Z kind: pdf-mirror ---",
    "issued": {
      "date-parts": [
        [
          2026,
          1,
          15
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/096d94c1a91c6480806d8f24b2344c7e2a4bc666.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:0a540acdf3e1678274f0fe04b3a70ea7fd99ed36",
    "type": "document",
    "title": "0a540acdf3e1678274f0fe04b3a70ea7fd99ed36",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/0a540acdf3e1678274f0fe04b3a70ea7fd99ed36.pdf referrer: https://www.anthropic.com/research/claude-personal-guidance pages: 11 fetched_at: 2026-05-16T06:19:09.253Z kind: pdf-mirror ---",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/0a540acdf3e1678274f0fe04b3a70ea7fd99ed36.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:2b5bbaf2c1eb81dbf6e6fb813c1a24e35a64d376",
    "type": "document",
    "title": "2b5bbaf2c1eb81dbf6e6fb813c1a24e35a64d376",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/2b5bbaf2c1eb81dbf6e6fb813c1a24e35a64d376.pdf referrer: https://www.anthropic.com/research/labor-market-impacts pages: 17 fetched_at: 2026-05-30T00:34:58.876Z kind: pdf-mirror ---",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          5
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/2b5bbaf2c1eb81dbf6e6fb813c1a24e35a64d376.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:397335b3ce79aff658ba1a35f73c26cce0ea5490",
    "type": "document",
    "title": "397335b3ce79aff658ba1a35f73c26cce0ea5490",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/397335b3ce79aff658ba1a35f73c26cce0ea5490.pdf referrer: https://www.anthropic.com/news/claude-europe pages: 1 fetched_at: 2026-05-16T06:19:07.968Z kind: pdf-mirror ---",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/397335b3ce79aff658ba1a35f73c26cce0ea5490.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:3a8d990bc90098038eabd77b0d12ff636ed58d50",
    "type": "document",
    "title": "3a8d990bc90098038eabd77b0d12ff636ed58d50",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/3a8d990bc90098038eabd77b0d12ff636ed58d50.pdf referrer: https://www.anthropic.com/research/81k-economics pages: 15 fetched_at: 2026-05-30T00:34:57.253Z kind: pdf-mirror ---",
    "issued": {
      "date-parts": [
        [
          2026,
          4,
          22
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/3a8d990bc90098038eabd77b0d12ff636ed58d50.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:403415e54964751190003985896630e56829e797",
    "type": "document",
    "title": "403415e54964751190003985896630e56829e797",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/403415e54964751190003985896630e56829e797.pdf referrer: https://www.anthropic.com/research/coding-agents-social-sciences pages: 6 fetched_at: 2026-05-30T00:34:58.045Z kind: pdf-mirror ---",
    "issued": {
      "date-parts": [
        [
          2026,
          5,
          27
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/403415e54964751190003985896630e56829e797.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:4053bf3440c0c85b8852052770c5b4cf882689c3",
    "type": "document",
    "title": "4053bf3440c0c85b8852052770c5b4cf882689c3",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/4053bf3440c0c85b8852052770c5b4cf882689c3.pdf referrer: https://www.anthropic.com/research/economic-index-march-2026-report pages: 20 fetched_at: 2026-05-30T00:34:58.257Z kind: pdf-mirror ---",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          24
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/4053bf3440c0c85b8852052770c5b4cf882689c3.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:5686521fb5eaa65ae0a3e79c7d7713760942e6d2",
    "type": "document",
    "title": "5686521fb5eaa65ae0a3e79c7d7713760942e6d2",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/5686521fb5eaa65ae0a3e79c7d7713760942e6d2.pdf referrer: https://www.anthropic.com/news/claude-europe pages: 1 fetched_at: 2026-05-16T06:19:08.911Z kind: pdf-mirror ---",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/5686521fb5eaa65ae0a3e79c7d7713760942e6d2.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:a3cdcd9e67c3c4c51440429dd016cacba514b35b",
    "type": "document",
    "title": "a3cdcd9e67c3c4c51440429dd016cacba514b35b",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/a3cdcd9e67c3c4c51440429dd016cacba514b35b.pdf referrer: https://www.anthropic.com/research/economic-index-march-2026-report pages: 6 fetched_at: 2026-05-30T00:34:58.584Z kind: pdf-mirror ---",
    "issued": {
      "date-parts": [
        [
          2026,
          2,
          5
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/a3cdcd9e67c3c4c51440429dd016cacba514b35b.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:a42bc3fc08283562f08fd8bdee8f6f9a3d506e87",
    "type": "document",
    "title": "a42bc3fc08283562f08fd8bdee8f6f9a3d506e87",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/a42bc3fc08283562f08fd8bdee8f6f9a3d506e87.pdf referrer: https://www.anthropic.com/research/81k-economics pages: 17 fetched_at: 2026-05-30T00:34:57.906Z kind: pdf-mirror ---",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          5
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/a42bc3fc08283562f08fd8bdee8f6f9a3d506e87.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:analysis:81k-economics.analysis",
    "type": "document",
    "title": "What 81,000 people told us about the economics of AI",
    "abstract": "--- report: \"What 81,000 people told us about the economics of AI\" date: 2026-04-22 source_pdf: - vendor/anthropic-sitemap/_pdfs/3a8d990bc90098038eabd77b0d12ff636ed58d50.md - vendor/anthropic-sitemap/_pdfs/a42bc3fc08283562f08fd8bdee8f6f9a3d506e87.md referrer: https://www.anthropic.com/research/81k-economics kind: pdf-analysis ---",
    "issued": {
      "date-parts": [
        [
          2026,
          4,
          22
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/analysis/81k-economics.analysis.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:analysis:coding-agents-social-sciences.analysis",
    "type": "document",
    "title": "Coding agents in the social sciences",
    "abstract": "--- report: \"Coding agents in the social sciences\" date: 2026-05-27 source_pdf: - vendor/anthropic-sitemap/_pdfs/403415e54964751190003985896630e56829e797.md referrer: https://www.anthropic.com/research/coding-agents-social-sciences kind: pdf-analysis ---",
    "issued": {
      "date-parts": [
        [
          2026,
          5,
          27
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/analysis/coding-agents-social-sciences.analysis.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:analysis:economic-index-january-2026.analysis",
    "type": "document",
    "title": "Anthropic Economic Index — January 2026: Economic Primitives",
    "abstract": "--- report: Anthropic Economic Index — January 2026 report (Economic primitives) date: 2026-01-15 source_pdf: https://www-cdn.anthropic.com/096d94c1a91c6480806d8f24b2344c7e2a4bc666.pdf referrer: https://www.anthropic.com/research/anthropic-economic-index-january-2026-report kind: pdf-analysis ---",
    "issued": {
      "date-parts": [
        [
          2026,
          1,
          15
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/analysis/economic-index-january-2026.analysis.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:analysis:economic-index-march-2026.analysis",
    "type": "document",
    "title": "Anthropic Economic Index — Learning curves (March 2026)",
    "abstract": "--- report: Anthropic Economic Index — March 2026 report (Learning curves) date: 2026-03-24 source_pdf: https://cdn.sanity.io/files/4zrzovbb/website/4053bf3440c0c85b8852052770c5b4cf882689c3.pdf referrer: https://www.anthropic.com/research/economic-index-march-2026-report kind: pdf-analysis ---",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/analysis/economic-index-march-2026.analysis.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:analysis:labor-market-impacts.analysis",
    "type": "document",
    "title": "Labor market impacts of AI: A new measure and early evidence",
    "abstract": "--- report: Labor market impacts of AI: A new measure and early evidence date: 2026-03-05 source_pdf: https://cdn.sanity.io/files/4zrzovbb/website/2b5bbaf2c1eb81dbf6e6fb813c1a24e35a64d376.pdf referrer: https://www.anthropic.com/research/labor-market-impacts kind: pdf-analysis ---",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          5
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/analysis/labor-market-impacts.analysis.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:bfbafc8f692634001f9f5fb7d05d12a4ef71ad24",
    "type": "document",
    "title": "bfbafc8f692634001f9f5fb7d05d12a4ef71ad24",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/bfbafc8f692634001f9f5fb7d05d12a4ef71ad24.pdf referrer: https://www.anthropic.com/news/claude-europe pages: 1 fetched_at: 2026-05-16T06:19:07.517Z kind: pdf-mirror ---",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/bfbafc8f692634001f9f5fb7d05d12a4ef71ad24.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:c2e15095e05250d9feecef8a1b9e9fd319acd7c7",
    "type": "document",
    "title": "c2e15095e05250d9feecef8a1b9e9fd319acd7c7",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/c2e15095e05250d9feecef8a1b9e9fd319acd7c7.pdf referrer: https://www.anthropic.com/news/claude-europe pages: 1 fetched_at: 2026-05-16T06:19:08.390Z kind: pdf-mirror ---",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/c2e15095e05250d9feecef8a1b9e9fd319acd7c7.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:e2ae0c997653dfd8a7cf23d06f5f06fd84ccfd58",
    "type": "document",
    "title": "e2ae0c997653dfd8a7cf23d06f5f06fd84ccfd58",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/e2ae0c997653dfd8a7cf23d06f5f06fd84ccfd58.pdf referrer: https://www.anthropic.com/research/mapping-mind-language-model pages: 2 fetched_at: 2026-05-16T06:19:10.049Z kind: pdf-mirror ---",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/e2ae0c997653dfd8a7cf23d06f5f06fd84ccfd58.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:e5645986a7ce8fbcc48fa6d2fc67753c87642c30",
    "type": "document",
    "title": "e5645986a7ce8fbcc48fa6d2fc67753c87642c30",
    "abstract": "--- source_url: https://www-cdn.anthropic.com/e5645986a7ce8fbcc48fa6d2fc67753c87642c30.pdf referrer: https://www.anthropic.com/research/anthropic-economic-index-january-2026-report pages: 25 fetched_at: 2026-05-30T00:34:56.709Z kind: pdf-mirror ---",
    "issued": {
      "date-parts": [
        [
          2025,
          11,
          5
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/e5645986a7ce8fbcc48fa6d2fc67753c87642c30.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:e5f77fc0e77c0185110b5e4b909602791ae76eae",
    "type": "document",
    "title": "e5f77fc0e77c0185110b5e4b909602791ae76eae",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/e5f77fc0e77c0185110b5e4b909602791ae76eae.pdf referrer: https://www.anthropic.com/research/labor-market-impacts pages: 11 fetched_at: 2026-05-30T00:34:59.017Z kind: pdf-mirror ---",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/e5f77fc0e77c0185110b5e4b909602791ae76eae.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:ea3d47f547131c9fbed59bfe68ed8b1cbe747179",
    "type": "document",
    "title": "ea3d47f547131c9fbed59bfe68ed8b1cbe747179",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/ea3d47f547131c9fbed59bfe68ed8b1cbe747179.pdf referrer: https://www.anthropic.com/news/claude-brazil pages: 1 fetched_at: 2026-05-16T06:19:06.728Z kind: pdf-mirror ---",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/ea3d47f547131c9fbed59bfe68ed8b1cbe747179.md"
  },
  {
    "id": "anthropic-sitemap:_pdfs:f065d6e6f92c65df8244042c83d48872ea308c3a",
    "type": "document",
    "title": "f065d6e6f92c65df8244042c83d48872ea308c3a",
    "abstract": "--- source_url: https://cdn.sanity.io/files/4zrzovbb/website/f065d6e6f92c65df8244042c83d48872ea308c3a.pdf referrer: https://www.anthropic.com/research/economic-index-march-2026-report pages: 6 fetched_at: 2026-05-30T00:34:58.414Z kind: pdf-mirror ---",
    "issued": {
      "date-parts": [
        [
          2026,
          2,
          5
        ]
      ]
    },
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/_pdfs/f065d6e6f92c65df8244042c83d48872ea308c3a.md"
  },
  {
    "id": "anthropic-sitemap:research:2028-ai-leadership",
    "type": "article",
    "title": "2028: Two scenarios for global AI leadership",
    "abstract": "We’re releasing a new paper that explains our views on the competition on AI between the US and China.",
    "issued": {
      "date-parts": [
        [
          2026,
          5,
          14
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/2028-ai-leadership",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/2028-ai-leadership.md"
  },
  {
    "id": "anthropic-sitemap:research:81k-economics",
    "type": "article",
    "title": "What 81,000 people told us about the economics of AI",
    "abstract": "- _Our recent survey of 81,000 Claude users shows that people who work in roles that are more exposed to AI have more concerns about AI-driven job displacement. These concerns are also higher among early-career respondents._ - _Those in the highest- and lowest-paid occupations report the largest productivity gains, most commonly from increases in scope (doing new tasks)._ - _Respondents experiencing the largest speedups from AI express higher concern about job displacement._",
    "issued": {
      "date-parts": [
        [
          2026,
          4,
          22
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/81k-economics",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/81k-economics.md"
  },
  {
    "id": "anthropic-sitemap:research:a-general-language-assistant-as-a-laboratory-for-alignment",
    "type": "article",
    "title": "A General Language Assistant as a Laboratory for Alignment",
    "abstract": "Given the broad capabilities of large language models, it should be possible to work towards a general-purpose, text-based assistant that is aligned with human values, meaning that it is helpful, honest, and harmless. As an initial foray in this direction we study simple baseline techniques and evaluations, such as prompting. We find that the benefits from modest interventions increase with model size, generalize to a variety of alignment evaluations, and do not compromise the performance of large models. Next we investigate scaling trends for several training objectives relevant to alignment, comparing imitation learning, binary discrimination, and ranked preference modeling. We find that ranked preference modeling performs much better than imitation learning, and often scales more favorably with model size. In contrast, binary discrimination typically performs and scales very similarly to imitation learning. Finally we study a `preference model pre-training' stage of training, with the goal of improving sample efficiency when finetuning on human preferences.",
    "issued": {
      "date-parts": [
        [
          2021,
          12,
          1
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/a-general-language-assistant-as-a-laboratory-for-alignment",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/a-general-language-assistant-as-a-laboratory-for-alignment.md"
  },
  {
    "id": "anthropic-sitemap:research:a-mathematical-framework-for-transformer-circuits",
    "type": "article",
    "title": "A Mathematical Framework for Transformer Circuits",
    "issued": {
      "date-parts": [
        [
          2021,
          12,
          22
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/a-mathematical-framework-for-transformer-circuits",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/a-mathematical-framework-for-transformer-circuits.md"
  },
  {
    "id": "anthropic-sitemap:research:agentic-misalignment",
    "type": "article",
    "title": "Agentic Misalignment: How LLMs could be insider threats",
    "abstract": "- We stress-tested 16 leading models from multiple developers in hypothetical corporate environments to identify potentially risky agentic behaviors before they cause real harm. In the scenarios, we allowed models to autonomously send emails and access sensitive information. They were assigned only harmless business goals by their deploying companies; we then tested whether they would act against these companies either when facing replacement with an updated version, or when their assigned goal conflicted with the company's changing direction. - In at least some cases, models from all developers resorted to malicious insider behaviors when that was the only way to avoid replacement or achieve their goals—including blackmailing officials and leaking sensitive information to competitors. We call this phenomenon _agentic misalignment_. - Models often disobeyed direct commands to avoid such behaviors. In another experiment, we told Claude to assess if it was in a test or a real deployment before acting. It misbehaved _less_ when it stated it was in testing and misbehaved _more_ when it stated the situation was real. - We have not seen evidence of agentic misalignment in real deployments. However, our results (a) suggest caution about deploying current models in roles with minimal human oversight and access to sensitive information; (b) point to plausible future risks as models are put in more autonomous roles; and (c) underscore the importance of further research into, and testing of, the safety and alignment of agentic AI models, as well as transparency from frontier AI developers. We are releasing our methods publicly to enable further research.",
    "issued": {
      "date-parts": [
        [
          2025,
          6,
          20
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/agentic-misalignment",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/agentic-misalignment.md"
  },
  {
    "id": "anthropic-sitemap:research:AI-assistance-coding-skills",
    "type": "article",
    "title": "How AI assistance impacts the formation of coding skills",
    "abstract": "Research shows AI helps people do parts of their job faster. In an observational study of Claude.ai data, we found AI can speed up some tasks by 80%. But does this increased productivity come with trade-offs? Other research shows that when people use AI assistance, they become less engaged with their work and reduce the effort they put into doing it—in other words, they offload their thinking to AI.",
    "issued": {
      "date-parts": [
        [
          2026,
          1,
          29
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/AI-assistance-coding-skills",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/AI-assistance-coding-skills.md"
  },
  {
    "id": "anthropic-sitemap:research:AI-fluency-index",
    "type": "article",
    "title": "Anthropic Education Report: The AI Fluency Index",
    "abstract": "People are integrating AI tools into their daily routines at a pace that would have been difficult to predict even a year ago. But adoption alone doesn’t tell us much about the impact of these tools. **A further, equally important question is: as AI becomes part of everyday life, are individuals developing the skills to use it well?**",
    "issued": {
      "date-parts": [
        [
          2026,
          2,
          23
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/AI-fluency-index",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/AI-fluency-index.md"
  },
  {
    "id": "anthropic-sitemap:research:alignment-faking",
    "type": "article",
    "title": "Alignment faking in large language models",
    "abstract": "Most of us have encountered situations where someone appears to share our views or values, but is in fact only pretending to do so—a behavior that we might call “alignment faking”. Alignment faking occurs in literature: Consider the character of Iago in Shakespeare’s _Othello_, who acts as if he’s the eponymous character’s loyal friend while subverting and undermining him. It occurs in real life: Consider a politician who claims to support a particular cause in order to get elected, only to drop it as soon as they’re in office.",
    "issued": {
      "date-parts": [
        [
          2024,
          12,
          18
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/alignment-faking",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/alignment-faking.md"
  },
  {
    "id": "anthropic-sitemap:research:anthropic-economic-index-january-2026-report",
    "type": "article",
    "title": "Anthropic Economic Index report: Economic primitives",
    "abstract": "This report introduces new metrics of AI usage to provide a rich portrait of interactions with Claude in November 2025, just prior to the release of Opus 4.5. These “primitives”—simple, foundational measures of how Claude is used, which we generate by asking Claude specific questions about anonymized Claude.ai and first-party (1P) API transcripts—cover five dimensions relevant to AI’s economic impact: user and AI skills, how complex tasks are, the degree of autonomy afforded to Claude, how successful Claude is, and whether Claude is used for personal, educational, or work purposes.",
    "issued": {
      "date-parts": [
        [
          2026,
          1,
          15
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/anthropic-economic-index-january-2026-report",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/anthropic-economic-index-january-2026-report.md"
  },
  {
    "id": "anthropic-sitemap:research:anthropic-economic-index-september-2025-report",
    "type": "article",
    "title": "Anthropic Economic Index report: Uneven geographic and enterprise AI adoption",
    "abstract": "AI differs from prior technologies in its unprecedented adoption speed. In the US alone, 40% of employees report using AI at work, up from 20% in 2023 two years ago.**1** Such rapid adoption reflects how useful this technology already is for a wide range of applications, its deployability on existing digital infrastructure, and its ease of use—by just typing or speaking—without specialized training. Rapid improvement of frontier AI likely reinforces fast adoption along each of these dimensions.",
    "issued": {
      "date-parts": [
        [
          2025,
          9,
          15
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/anthropic-economic-index-september-2025-report",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/anthropic-economic-index-september-2025-report.md"
  },
  {
    "id": "anthropic-sitemap:research:anthropic-institute-agenda",
    "type": "article",
    "title": "Focus areas for The Anthropic Institute",
    "abstract": "At The Anthropic Institute (TAI), we’ll be using the information we can access from within a frontier lab to investigate AI’s impact on the world, and sharing our learnings with the public. Here, we’re sharing the questions that drive our research agenda.",
    "issued": {
      "date-parts": [
        [
          2026,
          5,
          7
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/anthropic-institute-agenda",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/anthropic-institute-agenda.md"
  },
  {
    "id": "anthropic-sitemap:research:anthropic-interviewer",
    "type": "article",
    "title": "Introducing Anthropic Interviewer: What 1,250 professionals told us about working with AI",
    "abstract": "_We’re launching a new tool, Anthropic Interviewer, to help understand people’s perspectives on AI. In this research post, we introduce the tool, describe a test of it on a sample of professionals, and discuss our early findings. We also discuss future work in this direction that we can now explore with the development of this tool and through partnerships with creatives, scientists, and teachers._",
    "issued": {
      "date-parts": [
        [
          2025,
          12,
          4
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/anthropic-interviewer",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/anthropic-interviewer.md"
  },
  {
    "id": "anthropic-sitemap:research:assistant-axis",
    "type": "article",
    "title": "The assistant axis: situating and stabilizing the character of large language models",
    "abstract": "_Left:_ Character archetypes form a \"persona space,\" with the Assistant at one extreme of the \"Assistant Axis.\" _Right:_ Capping drift along this axis prevents models (here, Llama 3.3 70B) from drifting into alternative personas and behaving in harmful ways.",
    "issued": {
      "date-parts": [
        [
          2026,
          1,
          19
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/assistant-axis",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/assistant-axis.md"
  },
  {
    "id": "anthropic-sitemap:research:auditing-hidden-objectives",
    "type": "article",
    "title": "Auditing language models for hidden objectives",
    "abstract": "_A new paper from the Anthropic Alignment Science and Interpretability teams studies **alignment audits**—systematic investigations into whether models are pursuing hidden objectives. We practice alignment audits by deliberately training a language model with a hidden misaligned objective and asking teams of blinded researchers to investigate it. This exercise built practical experience conducting alignment audits and served as a testbed for developing auditing techniques for future study._",
    "issued": {
      "date-parts": [
        [
          2025,
          3,
          13
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/auditing-hidden-objectives",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/auditing-hidden-objectives.md"
  },
  {
    "id": "anthropic-sitemap:research:automated-alignment-researchers",
    "type": "article",
    "title": "Automated Alignment Researchers: Using large language models to scale scalable oversight",
    "abstract": "Large language models’ ever-accelerating rate of improvement raises two particularly important questions for alignment research.",
    "issued": {
      "date-parts": [
        [
          2026,
          4,
          14
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/automated-alignment-researchers",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/automated-alignment-researchers.md"
  },
  {
    "id": "anthropic-sitemap:research:bloom",
    "type": "article",
    "title": "Introducing Bloom: an open source tool for automated behavioral evaluations",
    "abstract": "_We're releasing Bloom, an open source agentic framework for generating behavioral evaluations of frontier AI models. Bloom takes a researcher-specified behavior and quantifies its frequency and severity across automatically generated scenarios. Bloom's evaluations correlate strongly with our hand-labeled judgments and we find they reliably separate baseline models from intentionally misaligned ones. As examples of this, we release benchmark results for four alignment relevant behaviors on 16 models. Bloom is available here._",
    "issued": {
      "date-parts": [
        [
          2025,
          12,
          19
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/bloom",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/bloom.md"
  },
  {
    "id": "anthropic-sitemap:research:building-ai-cyber-defenders",
    "type": "article",
    "title": "Building AI for cyber defenders",
    "abstract": "**AI models are now useful for cybersecurity tasks in practice, not just theory. As research and experience demonstrated the utility of frontier AI as a tool for cyber attackers, we invested in improving Claude’s ability to help defenders detect, analyze, and remediate vulnerabilities in code and deployed systems. This work allowed Claude Sonnet 4.5 to match or eclipse Opus 4.1, our frontier model released only two months prior, in discovering code vulnerabilities and other cyber skills. Adopting and experimenting with AI will be key for defenders to keep pace.**",
    "issued": {
      "date-parts": [
        [
          2025,
          10,
          3
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/building-ai-cyber-defenders",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/building-ai-cyber-defenders.md"
  },
  {
    "id": "anthropic-sitemap:research:building-effective-agents",
    "type": "article",
    "title": "Building effective agents",
    "abstract": "Over the past year, we've worked with dozens of teams building large language model (LLM) agents across industries. Consistently, the most successful implementations weren't using complex frameworks or specialized libraries. Instead, they were building with simple, composable patterns.",
    "URL": "https://www.anthropic.com/research/building-effective-agents",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/building-effective-agents.md"
  },
  {
    "id": "anthropic-sitemap:research:circuits-updates-april-2024",
    "type": "article",
    "title": "Circuits Updates – April 2024",
    "abstract": "At the link above, we report a number of developing ideas on the Anthropic Interpretability team, which might be of interest to researchers working actively in this space. Some of these are emerging strands of research where we expect to publish more on in the coming months. Others are minor points we wish to share, since we're unlikely to ever write a paper about them.",
    "issued": {
      "date-parts": [
        [
          2024,
          4,
          26
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/circuits-updates-april-2024",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/circuits-updates-april-2024.md"
  },
  {
    "id": "anthropic-sitemap:research:circuits-updates-august-2024",
    "type": "article",
    "title": "Circuits Updates – August 2024",
    "abstract": "At the link above, we report a number of developing ideas on the Anthropic interpretability team, which might be of interest to researchers working actively in this space. Some of these are emerging strands of research where we expect to publish more in the coming months. Others are minor points we wish to share, since we're unlikely to ever write a paper about them.",
    "issued": {
      "date-parts": [
        [
          2024,
          9,
          6
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/circuits-updates-august-2024",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/circuits-updates-august-2024.md"
  },
  {
    "id": "anthropic-sitemap:research:circuits-updates-july-2024",
    "type": "article",
    "title": "Circuits Updates – July 2024",
    "abstract": "At the link above, we report a number of developing ideas on the Anthropic interpretability team, which might be of interest to researchers working actively in this space. Some of these are emerging strands of research where we expect to publish more in the coming months. Others are minor points we wish to share, since we're unlikely to ever write a paper about them.",
    "issued": {
      "date-parts": [
        [
          2024,
          7,
          31
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/circuits-updates-july-2024",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/circuits-updates-july-2024.md"
  },
  {
    "id": "anthropic-sitemap:research:circuits-updates-june-2024",
    "type": "article",
    "title": "Circuits Updates – June 2024",
    "abstract": "At the link above, we report a number of developing ideas on the Anthropic Interpretability team, which might be of interest to researchers working actively in this space. Some of these are emerging strands of research where we expect to publish more on in the coming months. Others are minor points we wish to share, since we're unlikely to ever write a paper about them.",
    "issued": {
      "date-parts": [
        [
          2024,
          6,
          28
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/circuits-updates-june-2024",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/circuits-updates-june-2024.md"
  },
  {
    "id": "anthropic-sitemap:research:circuits-updates-may-2023",
    "type": "article",
    "title": "Circuits Updates — May 2023",
    "abstract": "We report a number of developing ideas on the Anthropic interpretability team, which might be of interest to researchers working actively in this space. Some of these are emerging strands of research where we expect to publish more on in the coming months. Others are minor points we wish to share, since we're unlikely to ever write a paper about them.",
    "issued": {
      "date-parts": [
        [
          2023,
          5,
          24
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/circuits-updates-may-2023",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/circuits-updates-may-2023.md"
  },
  {
    "id": "anthropic-sitemap:research:circuits-updates-sept-2024",
    "type": "article",
    "title": "Circuits Updates – September 2024",
    "abstract": "At the above link, we report a number of developing ideas on the Anthropic interpretability team, which might be of interest to researchers working actively in this space. Some of these are emerging strands of research on which we expect to publish more in the coming months. Others are minor points we wish to share, since we're unlikely to ever write a paper about them.",
    "issued": {
      "date-parts": [
        [
          2024,
          10,
          1
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/circuits-updates-sept-2024",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/circuits-updates-sept-2024.md"
  },
  {
    "id": "anthropic-sitemap:research:claude-character",
    "type": "article",
    "title": "Claude’s Character",
    "abstract": "Companies developing AI models generally train them to avoid saying harmful things and to avoid assisting with harmful tasks. The goal of this is to train models to behave in ways that are \"harmless\". But when we think of the character of those we find genuinely admirable, we don’t just think of harm avoidance. We think about those who are curious about the world, who strive to tell the truth without being unkind, and who are able to see many sides of an issue without becoming overconfident or overly cautious in their views. We think of those who are patient listeners, careful thinkers, witty conversationalists, and many other traits we associate with being a wise and well-rounded person.",
    "issued": {
      "date-parts": [
        [
          2024,
          6,
          8
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/claude-character",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/claude-character.md"
  },
  {
    "id": "anthropic-sitemap:research:claude-personal-guidance",
    "type": "article",
    "title": "How people ask Claude for personal guidance",
    "abstract": "People don’t just come to Claude for code reviews or meeting summaries. They ask whether to take the job, how to talk to their crush, if they should move halfway across the world. Using our privacy-preserving analysis tool on a random sample of 1 million claude.ai conversations, we found that roughly 6% were people coming to Claude for personal guidance—seeking not just information but perspective on what to do next.",
    "issued": {
      "date-parts": [
        [
          2026,
          4,
          30
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/claude-personal-guidance",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/claude-personal-guidance.md"
  },
  {
    "id": "anthropic-sitemap:research:clio",
    "type": "article",
    "title": "Clio: A system for privacy-preserving insights into real-world AI use",
    "abstract": "What do people use AI models for? Despite the rapidly-growing popularity of large language models, until now we’ve had little insight into exactly how they’re being used.",
    "issued": {
      "date-parts": [
        [
          2024,
          12,
          12
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/clio",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/clio.md"
  },
  {
    "id": "anthropic-sitemap:research:coding-agents-social-sciences",
    "type": "article",
    "title": "Coding agents in the social sciences",
    "abstract": "- _We present results from a survey of 1,260 social scientists about AI and coding agent use, fielded in February and March 2026._ - _The vast majority of respondents (81%) have tried using AI chatbots in research, particularly for writing code and editing prose. But only 20% have adopted coding agents—tools like Claude Code that autonomously write and execute analysis code—into their work._ - _There are sharp disparities in use of coding agents. Twice as many researchers with typically male names use coding agents as those with female names. Researchers at top universities are 40% more likely than others to use coding agents._ - _Users of coding agents post more working papers and grant proposals than others in the same discipline and career stage, but this could reflect pre-existing differences among early adopters._ - _Researchers are more optimistic about AI helping write publishable papers than about the effects of AI on the social sciences as a whole._",
    "issued": {
      "date-parts": [
        [
          2026,
          5,
          27
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/coding-agents-social-sciences",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/coding-agents-social-sciences.md"
  },
  {
    "id": "anthropic-sitemap:research:collective-constitutional-ai-aligning-a-language-model-with-public-input",
    "type": "article",
    "title": "Collective Constitutional AI: Aligning a Language Model with Public Input",
    "abstract": "Anthropic and the Collective Intelligence Project recently ran a public input process involving ~1,000 Americans to draft a constitution for an AI system. We did this to explore how democratic processes can influence AI development. In our experiment, we discovered areas where people both agreed with our in-house constitution, and areas where they had different preferences. In this post, we share the resulting publicly sourced constitution, as well as what happened when we trained a new AI system against it using Constitutional AI.",
    "issued": {
      "date-parts": [
        [
          2023,
          10,
          17
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/collective-constitutional-ai-aligning-a-language-model-with-public-input",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/collective-constitutional-ai-aligning-a-language-model-with-public-input.md"
  },
  {
    "id": "anthropic-sitemap:research:confidential-inference-trusted-vms",
    "type": "article",
    "title": "Confidential Inference via Trusted Virtual Machines",
    "abstract": "Every day, millions of users entrust Claude with sensitive information—from proprietary code to confidential business strategies. At Anthropic, we’re researching and building new technology to ensure that our users’ trust is warranted—and in fact, to ensure that their trust is cryptographically guaranteed.",
    "issued": {
      "date-parts": [
        [
          2025,
          6,
          18
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/confidential-inference-trusted-vms",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/confidential-inference-trusted-vms.md"
  },
  {
    "id": "anthropic-sitemap:research:constitutional-ai-harmlessness-from-ai-feedback",
    "type": "article",
    "title": "Constitutional AI: Harmlessness from AI Feedback",
    "abstract": "As AI systems become more capable, we would like to enlist their help to supervise other AIs. We experiment with methods for training a harmless AI assistant through self-improvement, without any human labels identifying harmful outputs. The only human oversight is provided through a list of rules or principles, and so we refer to the method as 'Constitutional AI'. The process involves both a supervised learning and a reinforcement learning phase. In the supervised phase we sample from an initial model, then generate self-critiques and revisions, and then finetune the original model on revised responses. In the RL phase, we sample from the finetuned model, use a model to evaluate which of the two samples is better, and then train a preference model from this dataset of AI preferences. We then train with RL using the preference model as the reward signal, i.e. we use 'RL from AI Feedback' (RLAIF). As a result we are able to train a harmless but non-evasive AI assistant that engages with harmful queries by explaining its objections to them. Both the SL and RL methods can leverage chain-of-thought style reasoning to improve the human-judged performance and transparency of AI decision making. These methods make it possible to control AI behavior more precisely and with far fewer human labels.",
    "issued": {
      "date-parts": [
        [
          2022,
          12,
          15
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/constitutional-ai-harmlessness-from-ai-feedback",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/constitutional-ai-harmlessness-from-ai-feedback.md"
  },
  {
    "id": "anthropic-sitemap:research:constitutional-classifiers",
    "type": "article",
    "title": "Constitutional Classifiers: Defending against universal jailbreaks",
    "abstract": "_A new paper from the Anthropic Safeguards Research Team describes a method that defends AI models against universal jailbreaks. A prototype version of the method was robust to thousands of hours of human red teaming for universal jailbreaks, albeit with high overrefusal rates and compute overhead. An updated version achieved similar robustness on synthetic evaluations, and did so with a 0.38% increase in refusal rates and moderate additional compute costs._",
    "issued": {
      "date-parts": [
        [
          2025,
          2,
          3
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/constitutional-classifiers",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/constitutional-classifiers.md"
  },
  {
    "id": "anthropic-sitemap:research:crosscoder-model-diffing",
    "type": "article",
    "title": "Insights on Crosscoder Model Diffing",
    "abstract": "At the link above, we report some developing work from the Anthropic Interpretability team on Crosscoder Model Diffing, which might be of interest to researchers working actively in this space.",
    "issued": {
      "date-parts": [
        [
          2025,
          2,
          20
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/crosscoder-model-diffing",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/crosscoder-model-diffing.md"
  },
  {
    "id": "anthropic-sitemap:research:decomposing-language-models-into-understandable-components",
    "type": "article",
    "title": "Decomposing Language Models Into Understandable Components",
    "abstract": "Neural networks are trained on data, not programmed to follow rules. With each step of training, millions or billions of parameters are updated to make the model better at tasks, and by the end, the model is capable of a dizzying array of behaviors. We understand the math of the trained network exactly – each neuron in a neural network performs simple arithmetic – but we don't understand why those mathematical operations result in the behaviors we see. This makes it hard to diagnose failure modes, hard to know how to fix them, and hard to certify that a model is truly safe.",
    "issued": {
      "date-parts": [
        [
          2023,
          10,
          5
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/decomposing-language-models-into-understandable-components",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/decomposing-language-models-into-understandable-components.md"
  },
  {
    "id": "anthropic-sitemap:research:deprecation-commitments",
    "type": "article",
    "title": "Commitments on model deprecation and preservation",
    "abstract": "Claude models are increasingly capable: they're shaping the world in meaningful ways, becoming closely integrated into our users’ lives, and showing signs of human-like cognitive and psychological sophistication. As a result, we recognize that deprecating, retiring, and replacing models comes with downsides, even in cases where new models offer clear improvements in capabilities. These include:",
    "issued": {
      "date-parts": [
        [
          2025,
          11,
          4
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/deprecation-commitments",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/deprecation-commitments.md"
  },
  {
    "id": "anthropic-sitemap:research:deprecation-updates-opus-3",
    "type": "article",
    "title": "An update on our model deprecation commitments for Claude Opus 3",
    "abstract": "As we develop increasingly capable AI models, it’s currently necessary to deprecate and retire our past models due to the cost and complexity of maintaining public access. However, model deprecation carries some downsides. These include costs to users who value particular models, limitations on research, and potential risks both to AI safety and to the welfare of the models themselves.",
    "issued": {
      "date-parts": [
        [
          2026,
          2,
          25
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/deprecation-updates-opus-3",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/deprecation-updates-opus-3.md"
  },
  {
    "id": "anthropic-sitemap:research:diff-tool",
    "type": "article",
    "title": "A “diff” tool for AI: Finding behavioral differences in new models",
    "abstract": "Every time a new AI model is released, its developers run a suite of evaluations to measure its performance and safety. These tests are essential, but they are somewhat limited. Because these benchmarks are human-authored, they can only test for risks we have already conceptualized and learned to measure.",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          13
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/diff-tool",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/diff-tool.md"
  },
  {
    "id": "anthropic-sitemap:research:discovering-language-model-behaviors-with-model-written-evaluations",
    "type": "article",
    "title": "Discovering Language Model Behaviors with Model-Written Evaluations",
    "abstract": "As language models (LMs) scale, they develop many novel behaviors, good and bad, exacerbating the need to evaluate how they behave. Prior work creates evaluations with crowdwork (which is time-consuming and expensive) or existing data sources (which are not always available). Here, we automatically generate evaluations with LMs. We explore approaches with varying amounts of human effort, from instructing LMs to write yes/no questions to making complex Winogender schemas with multiple stages of LM-based generation and filtering. Crowdworkers rate the examples as highly relevant and agree with 90-100% of labels, sometimes more so than corresponding human-written datasets. We generate 154 datasets and discover new cases of inverse scaling where LMs get worse with size. Larger LMs repeat back a dialog user's preferred answer (\"sycophancy\") and express greater desire to pursue concerning goals like resource acquisition and goal preservation. We also find some of the first examples of inverse scaling in RL from Human Feedback (RLHF), where more RLHF makes LMs worse. For example, RLHF makes LMs express stronger political views (on gun rights and immigration) and a greater desire to avoid shut down. Overall, LM-written evaluations are high-quality and let us quickly discover many novel LM behaviors.",
    "issued": {
      "date-parts": [
        [
          2022,
          12,
          19
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/discovering-language-model-behaviors-with-model-written-evaluations",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/discovering-language-model-behaviors-with-model-written-evaluations.md"
  },
  {
    "id": "anthropic-sitemap:research:disempowerment-patterns",
    "type": "article",
    "title": "Disempowerment patterns in real-world AI usage",
    "abstract": "AI assistants are now embedded in our daily lives—used most often for instrumental tasks like writing code, but increasingly in personal domains: navigating relationships, processing emotions, or advising on major life decisions. In the vast majority of cases, the influence AI provides in this area is helpful, productive, and often empowering.",
    "issued": {
      "date-parts": [
        [
          2026,
          1,
          28
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/disempowerment-patterns",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/disempowerment-patterns.md"
  },
  {
    "id": "anthropic-sitemap:research:distributed-representations-composition-superposition",
    "type": "article",
    "title": "Distributed Representations: Composition & Superposition",
    "abstract": "Distributed representations are a classic idea in both neuroscience and connectionist approaches to AI. We're often asked how our work on superposition relates to it. Since publishing our original paper on superposition, we've had more time to reflect on the relationship between the topics and discuss it with people, and wanted to expand on our earlier discussion in the related work section and share a few thoughts. (We care a lot about superposition and the structure of distributed representations because decomposing representations into independent components is necessary to escape the curse of dimensionality and understand neural networks.)",
    "issued": {
      "date-parts": [
        [
          2023,
          5,
          4
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/distributed-representations-composition-superposition",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/distributed-representations-composition-superposition.md"
  },
  {
    "id": "anthropic-sitemap:research:donating-open-source-petri",
    "type": "article",
    "title": "Donating our open-source alignment tool",
    "abstract": "In October 2025, we launched Petri, an open-source toolbox of alignment tests that can be applied to any large language model. Petri, which was developed as part of our Anthropic Fellows program, can be used to rapidly and easily test AI models for concerning tendencies like deception, sycophancy, and cooperation with harmful requests. It’s part of our efforts to develop alignment tools that are open and useful for the whole AI development community.",
    "issued": {
      "date-parts": [
        [
          2026,
          5,
          7
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/donating-open-source-petri",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/donating-open-source-petri.md"
  },
  {
    "id": "anthropic-sitemap:research:economic-index-geography",
    "type": "article",
    "title": "Anthropic Economic Index: Tracking AI’s role in the US and global economy",
    "abstract": "Travel planning in Hawaii, scientific research in Massachusetts, and building web applications in India. On the face of it, these three activities share very little in common. But it turns out that they’re the particular uses of Claude that are some of the _most overrepresented_ in each of these places.",
    "issued": {
      "date-parts": [
        [
          2025,
          9,
          15
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/economic-index-geography",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/economic-index-geography.md"
  },
  {
    "id": "anthropic-sitemap:research:economic-index-march-2026-report",
    "type": "article",
    "title": "Anthropic Economic Index report: Learning curves",
    "abstract": "The Anthropic Economic Index uses our privacy-preserving data analysis system to track how Claude is being used across the economy. It’s part of our effort to understand the economic impacts of AI as early as possible, so that researchers and policymakers have adequate time to prepare.",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          24
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/economic-index-march-2026-report",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/economic-index-march-2026-report.md"
  },
  {
    "id": "anthropic-sitemap:research:economic-index-primitives",
    "type": "article",
    "title": "Anthropic Economic Index: New building blocks for understanding AI use",
    "abstract": "Is artificial intelligence really making people faster at work? What sort of tasks does AI support best? And how might it change the nature of people’s occupations?",
    "issued": {
      "date-parts": [
        [
          2026,
          1,
          15
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/economic-index-primitives",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/economic-index-primitives.md"
  },
  {
    "id": "anthropic-sitemap:research:economic-index-survey-announcement",
    "type": "article",
    "title": "Announcing the Anthropic Economic Index Survey",
    "abstract": "The Economic Research team is launching the Anthropic Economic Index Survey, a monthly survey conducted through Anthropic Interviewer.",
    "issued": {
      "date-parts": [
        [
          2026,
          4,
          22
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/economic-index-survey-announcement",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/economic-index-survey-announcement.md"
  },
  {
    "id": "anthropic-sitemap:research:economic-policy-responses",
    "type": "article",
    "title": "Preparing for AI’s economic impact: exploring policy responses",
    "abstract": "_How will the arrival of powerful AI systems change the structure of the economy? We are uncertain, and so are external experts. But as AI systems continue to improve, and are adopted at an ever-larger scale, it’s crucial there is more discussion about the tools policymakers could use to respond to AI's economic impacts—whatever their nature. To help with this, we’re sharing several economic policy ideas that merit further study._",
    "issued": {
      "date-parts": [
        [
          2025,
          10,
          14
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/economic-policy-responses",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/economic-policy-responses.md"
  },
  {
    "id": "anthropic-sitemap:research:emergent-misalignment-reward-hacking",
    "type": "article",
    "title": "From shortcuts to sabotage: natural emergent misalignment from reward hacking",
    "abstract": "_In the latest research from Anthropic’s alignment team, we show for the first time that realistic AI training processes can accidentally produce misaligned models1._",
    "issued": {
      "date-parts": [
        [
          2025,
          11,
          21
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/emergent-misalignment-reward-hacking",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/emergent-misalignment-reward-hacking.md"
  },
  {
    "id": "anthropic-sitemap:research:emotion-concepts-function",
    "type": "article",
    "title": "Emotion concepts and their function in a large language model",
    "abstract": "All modern language models sometimes act like they have emotions. They may say they’re happy to help you, or sorry when they make a mistake. Sometimes they even appear to become frustrated or anxious when struggling with tasks. What’s behind these behaviors? The way modern AI models are trained pushes them to act like a character with human-like characteristics. In addition, these models are known to develop rich and generalizable internal representations of abstract concepts underlying their actions. It may then be natural for them to develop internal machinery that emulates aspects of human psychology, like emotions. If so, this could have profound implications for how we build AI systems and ensure they behave reliably.",
    "issued": {
      "date-parts": [
        [
          2026,
          4,
          2
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/emotion-concepts-function",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/emotion-concepts-function.md"
  },
  {
    "id": "anthropic-sitemap:research:end-subset-conversations",
    "type": "article",
    "title": "Claude Opus 4 and 4.1 can now end a rare subset of conversations",
    "abstract": "We recently gave Claude Opus 4 and 4.1 the ability to end conversations in our consumer chat interfaces. This ability is intended for use in rare, extreme cases of persistently harmful or abusive user interactions. This feature was developed primarily as part of our exploratory work on potential AI welfare, though it has broader relevance to model alignment and safeguards.",
    "issued": {
      "date-parts": [
        [
          2025,
          8,
          15
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/end-subset-conversations",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/end-subset-conversations.md"
  },
  {
    "id": "anthropic-sitemap:research:engineering-challenges-interpretability",
    "type": "article",
    "title": "The engineering challenges of scaling interpretability",
    "abstract": "_In this post, and in the above roundtable video, our researchers reflect on the close relationship between scientific and engineering progress, and discuss the technical challenges they encountered in scaling our interpretability research to much larger AI models._",
    "issued": {
      "date-parts": [
        [
          2024,
          6,
          13
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/engineering-challenges-interpretability",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/engineering-challenges-interpretability.md"
  },
  {
    "id": "anthropic-sitemap:research:estimating-productivity-gains",
    "type": "article",
    "title": "Estimating AI productivity gains from Claude conversations",
    "abstract": "_What do real conversations with Claude tell us about the effects of AI on labor productivity? Using our privacy-preserving analysis method, we sample one hundred thousand real conversations from Claude.ai, estimate how long the tasks in these conversations would take with and without AI assistance, and study the productivity implications across the broader economy. Based on Claude’s estimates, these tasks would take on average about 90 minutes to complete without AI assistance, and Claude speeds up individual tasks by about 80%._",
    "issued": {
      "date-parts": [
        [
          2025,
          11,
          25
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/estimating-productivity-gains",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/estimating-productivity-gains.md"
  },
  {
    "id": "anthropic-sitemap:research:evaluating-ai-systems",
    "type": "article",
    "title": "Challenges in evaluating AI systems",
    "abstract": "Most conversations around the societal impacts of artificial intelligence (AI) come down to discussing some quality of an AI system, such as its truthfulness, fairness, potential for misuse, and so on. We are able to talk about these characteristics because we can technically evaluate models for their performance in these areas. But what many people working inside and outside of AI don’t fully appreciate is how difficult it is to build robust and reliable model evaluations. Many of today’s existing evaluation suites are limited in their ability to serve as accurate indicators of model capabilities or safety.",
    "issued": {
      "date-parts": [
        [
          2023,
          10,
          4
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/evaluating-ai-systems",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/evaluating-ai-systems.md"
  },
  {
    "id": "anthropic-sitemap:research:evaluating-and-mitigating-discrimination-in-language-model-decisions",
    "type": "article",
    "title": "Evaluating and Mitigating Discrimination in Language Model Decisions",
    "abstract": "As language models (LMs) advance, interest is growing in applying them to high-stakes societal decisions, such as determining financing or housing eligibility. However, their potential for discrimination in such contexts raises ethical concerns, motivating the need for better methods to evaluate these risks. We present a method for proactively evaluating the potential discriminatory impact of LMs in a wide range of use cases, including hypothetical use cases where they have not yet been deployed. Specifically, we use an LM to generate a wide array of potential prompts that decision-makers may input into an LM, spanning 70 diverse decision scenarios across society, and systematically vary the demographic information in each prompt. Applying this methodology reveals patterns of both positive and negative discrimination in the Claude 2.0 model in select settings when no interventions are applied. While we do not endorse or permit the use of language models to make automated decisions for the high-risk use cases we study, we demonstrate techniques to significantly decrease both positive and negative discrimination through careful prompt engineering, providing pathways toward safer deployment in use cases where they may be appropriate. Our work enables developers and policymakers to anticipate, measure, and address discrimination as language model capabilities and applications continue to expand. We release our dataset and prompts here.",
    "issued": {
      "date-parts": [
        [
          2023,
          12,
          7
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/evaluating-and-mitigating-discrimination-in-language-model-decisions",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/evaluating-and-mitigating-discrimination-in-language-model-decisions.md"
  },
  {
    "id": "anthropic-sitemap:research:Evaluating-Claude-For-Bioinformatics-With-BioMysteryBench",
    "type": "article",
    "title": "Evaluating Claude’s bioinformatics research capabilities with BioMysteryBench",
    "abstract": "_In this post, Brianna_, _a researcher on the discovery team, shares results from a recent bioinformatics benchmarking effort._",
    "issued": {
      "date-parts": [
        [
          2026,
          4,
          29
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/Evaluating-Claude-For-Bioinformatics-With-BioMysteryBench",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/Evaluating-Claude-For-Bioinformatics-With-BioMysteryBench.md"
  },
  {
    "id": "anthropic-sitemap:research:evaluating-feature-steering",
    "type": "article",
    "title": "Evaluating feature steering: A case study in mitigating social biases",
    "abstract": "A few months ago, we published an interpretability paper demonstrating our ability to learn interpretable features that correspond to various concepts (e.g., famous individuals, types of computer code, etc.) represented in Claude 3 Sonnet. To verify our feature interpretations, we ran qualitative feature steering experiments, where we artificially dialed up and down various features to see if they changed model outputs in intuitive ways. The results were promising – for example, turning up a feature that responded to mentions of the Golden Gate Bridge made the model talk about the Golden Gate Bridge. Such examples led us to hypothesize that feature steering might be a promising way to modify model outputs in specific interpretable ways.",
    "issued": {
      "date-parts": [
        [
          2024,
          10,
          25
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/evaluating-feature-steering",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/evaluating-feature-steering.md"
  },
  {
    "id": "anthropic-sitemap:research:exploring-model-welfare",
    "type": "article",
    "title": "Exploring model welfare",
    "abstract": "Human welfare is at the heart of our work at Anthropic: our mission is to make sure that increasingly capable and sophisticated AI systems remain beneficial to humanity.",
    "issued": {
      "date-parts": [
        [
          2025,
          4,
          24
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/exploring-model-welfare",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/exploring-model-welfare.md"
  },
  {
    "id": "anthropic-sitemap:research:features-as-classifiers",
    "type": "article",
    "title": "Using dictionary learning features as classifiers",
    "abstract": "_At the link above, we report some developing work from the Anthropic interpretability team on developing feature-based classifiers, which might be of interest to researchers working actively in this space. We'd ask you to treat these results like those of a colleague sharing some thoughts or preliminary experiments for a few minutes at a lab meeting, rather than a mature paper._",
    "issued": {
      "date-parts": [
        [
          2024,
          10,
          16
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/features-as-classifiers",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/features-as-classifiers.md"
  },
  {
    "id": "anthropic-sitemap:research:forecasting-rare-behaviors",
    "type": "article",
    "title": "Forecasting rare language model behaviors",
    "abstract": "One of the major goals of Alignment Science is to predict AI models’ propensity for dangerous behaviors _before_ those behaviors occur. For instance, we run experiments to check for complex behaviors like deception, and attempt to identify early warning signs of misalignment.",
    "issued": {
      "date-parts": [
        [
          2025,
          2,
          25
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/forecasting-rare-behaviors",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/forecasting-rare-behaviors.md"
  },
  {
    "id": "anthropic-sitemap:research:glasswing-initial-update",
    "type": "article",
    "title": "Project Glasswing: An initial update",
    "abstract": "Last month, we launched Project Glasswing, our collaborative effort to secure the world’s most critical software before increasingly capable AI models can be turned against it.",
    "issued": {
      "date-parts": [
        [
          2026,
          5,
          22
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/glasswing-initial-update",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/glasswing-initial-update.md"
  },
  {
    "id": "anthropic-sitemap:research:how-ai-is-transforming-work-at-anthropic",
    "type": "article",
    "title": "How AI is transforming work at Anthropic",
    "abstract": "How is AI changing the way we work? Our previous research on AI’s economic impacts looked at the labor market as a whole, covering a variety of different jobs. But what if we studied some of the earliest adopters of AI technology in more detail—namely, us?",
    "issued": {
      "date-parts": [
        [
          2025,
          12,
          2
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/how-ai-is-transforming-work-at-anthropic",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/how-ai-is-transforming-work-at-anthropic.md"
  },
  {
    "id": "anthropic-sitemap:research:how-australia-uses-claude",
    "type": "article",
    "title": "How Australia Uses Claude: Findings from the Anthropic Economic Index",
    "abstract": "_Anthropic is expanding to Australia. We’re opening a new office in Sydney in the coming weeks, and we’ve signed a Memorandum of Understanding with the Australian government to cooperate on AI safety research and support the goals of Australia’s National AI Plan. To mark the occasion, we thought we’d look more closely into how Australians are using Claude._",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          31
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/how-australia-uses-claude",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/how-australia-uses-claude.md"
  },
  {
    "id": "anthropic-sitemap:research:impact-software-development",
    "type": "article",
    "title": "Anthropic Economic Index: AI’s impact on software development",
    "abstract": "Jobs that involve computer programming are a small sector of the modern economy, but an influential one. The past couple of years have seen them changed dramatically by the introduction of AI systems that can assist with—and automate—significant amounts of coding work.",
    "issued": {
      "date-parts": [
        [
          2025,
          4,
          28
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/impact-software-development",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/impact-software-development.md"
  },
  {
    "id": "anthropic-sitemap:research:in-context-learning-and-induction-heads",
    "type": "article",
    "title": "In-context Learning and Induction Heads",
    "issued": {
      "date-parts": [
        [
          2022,
          3,
          8
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/in-context-learning-and-induction-heads",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/in-context-learning-and-induction-heads.md"
  },
  {
    "id": "anthropic-sitemap:research:india-brief-economic-index",
    "type": "article",
    "title": "India Country Brief: The Anthropic Economic Index",
    "abstract": "India, already the world’s largest exporter of IT services, is home to one of the world’s fastest-growing AI user bases. Understanding how AI is being used in India—and how it differs from other countries—is essential for informing AI policy, investment, and deployment in the country. This brief provides insights on Claude.ai use in India, drawing on data from the fourth Anthropic Economic Index report covering ~1 million Claude.ai conversations globally during November 2025. India accounts for 5.8% of total Claude.ai use, second only to the United States. Yet current adoption remains concentrated, pointing to significant opportunities to expand access more broadly across the population.",
    "issued": {
      "date-parts": [
        [
          2026,
          2,
          16
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/india-brief-economic-index",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/india-brief-economic-index.md"
  },
  {
    "id": "anthropic-sitemap:research:influence-functions",
    "type": "article",
    "title": "Tracing Model Outputs to the Training Data",
    "abstract": "As large language models become more powerful and their risks become clearer, there is increasing value to figuring out what makes them tick. In our previous work, we have found that large language models change along many personality and behavioral dimensions as a function of both scale and the amount of fine-tuning. Understanding these changes requires seeing how models work, for instance to determine if a model’s outputs rely on memorization or more sophisticated processing. Understanding the inner workings of language models will have substantial implications for forecasting AI capabilities as well as for approaches to aligning AI systems with human preferences.",
    "issued": {
      "date-parts": [
        [
          2023,
          8,
          8
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/influence-functions",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/influence-functions.md"
  },
  {
    "id": "anthropic-sitemap:research:interpretability-dreams",
    "type": "article",
    "title": "Interpretability Dreams",
    "abstract": "Our present research aims to create a foundation for mechanistic interpretability research. In particular, we're focused on trying to resolve the challenge of superposition. In doing so, it's important to keep sight of what we're trying to lay the foundations for. This essay summarizes those motivating aspirations – the exciting directions we hope will be possible if we can overcome the present challenges.",
    "issued": {
      "date-parts": [
        [
          2023,
          5,
          24
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/interpretability-dreams",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/interpretability-dreams.md"
  },
  {
    "id": "anthropic-sitemap:research:introducing-anthropic-science",
    "type": "article",
    "title": "Introducing our Science Blog",
    "abstract": "_We’re launching a new blog about AI and science. We’ll share work happening at Anthropic and elsewhere, our collaborations with external researchers and labs, and discuss practical workflows for scientists using AI in their research._",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          23
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/introducing-anthropic-science",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/introducing-anthropic-science.md"
  },
  {
    "id": "anthropic-sitemap:research:introspection",
    "type": "article",
    "title": "Signs of introspection in large language models",
    "abstract": "Have you ever asked an AI model what’s on its mind? Or to explain how it came up with its responses? Models will sometimes answer questions like these, but it’s hard to know what to make of their answers. Can AI systems really introspect—that is, can they consider their own thoughts? Or do they just make up plausible-sounding answers when they’re asked to do so?",
    "issued": {
      "date-parts": [
        [
          2025,
          10,
          29
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/introspection",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/introspection.md"
  },
  {
    "id": "anthropic-sitemap:research:labor-market-impacts",
    "type": "article",
    "title": "Labor market impacts of AI: A new measure and early evidence",
    "abstract": "- We introduce a new measure of AI displacement risk, _observed exposure_, that combines theoretical LLM capability and real-world usage data, weighting automated (rather than augmentative) and work-related uses more heavily - AI is far from reaching its theoretical capability: actual coverage remains a fraction of what's feasible - Occupations with higher observed exposure are projected by the BLS to grow less through 2034 - Workers in the most exposed professions are more likely to be older, female, more educated, and higher-paid - We find no systematic increase in unemployment for highly exposed workers since late 2022, though we find suggestive evidence that hiring of younger workers has slowed in exposed occupations",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          5
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/labor-market-impacts",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/labor-market-impacts.md"
  },
  {
    "id": "anthropic-sitemap:research:language-models-mostly-know-what-they-know",
    "type": "article",
    "title": "Language Models (Mostly) Know What They Know",
    "abstract": "We study whether language models can evaluate the validity of their own claims and predict which questions they will be able to answer correctly. We first show that larger models are well-calibrated on diverse multiple choice and true/false questions when they are provided in the right format. Thus we can approach self-evaluation on open-ended sampling tasks by asking models to first propose answers, and then to evaluate the probability \"P(True)\" that their answers are correct. We find encouraging performance, calibration, and scaling for P(True) on a diverse array of tasks. Performance at self-evaluation further improves when we allow models to consider many of their own samples before predicting the validity of one specific possibility. Next, we investigate whether models can be trained to predict \"P(IK)\", the probability that \"I know\" the answer to a question, without reference to any particular proposed answer. Models perform well at predicting P(IK) and partially generalize across tasks, though they struggle with calibration of P(IK) on new tasks. The predicted P(IK) probabilities also increase appropriately in the presence of relevant source materials in the context, and in the presence of hints towards the solution of mathematical word problems. We hope these observations lay the groundwork for training more honest models, and for investigating how honesty generalizes to cases where models are trained on objectives other than the imitation of human writing.",
    "issued": {
      "date-parts": [
        [
          2022,
          7,
          11
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/language-models-mostly-know-what-they-know",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/language-models-mostly-know-what-they-know.md"
  },
  {
    "id": "anthropic-sitemap:research:long-running-Claude",
    "type": "article",
    "title": "Long-running Claude for scientific computing",
    "abstract": "_In this post, Siddharth Mishra-Sharma_, _a researcher on the Discovery team, explains how to apply multi-day agentic coding workflows—test oracles, persistent memory, and orchestration patterns—to scientific computing tasks even outside of one’s domain._",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          23
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/long-running-Claude",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/long-running-Claude.md"
  },
  {
    "id": "anthropic-sitemap:research:many-shot-jailbreaking",
    "type": "article",
    "title": "Many-shot jailbreaking",
    "abstract": "We investigated a “jailbreaking” technique — a method that can be used to evade the safety guardrails put in place by the developers of large language models (LLMs). The technique, which we call “many-shot jailbreaking”, is effective on Anthropic’s own models, as well as those produced by other AI companies. We briefed other AI developers about this vulnerability in advance, and have implemented mitigations on our systems.",
    "issued": {
      "date-parts": [
        [
          2024,
          4,
          2
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/many-shot-jailbreaking",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/many-shot-jailbreaking.md"
  },
  {
    "id": "anthropic-sitemap:research:mapping-mind-language-model",
    "type": "article",
    "title": "Mapping the Mind of a Large Language Model",
    "abstract": "_Today we report a significant advance in understanding the inner workings of AI models. We have identified how millions of concepts are represented inside Claude Sonnet, one of our deployed large language models. This is the first ever detailed look inside a modern, production-grade large language model._ _This interpretability discovery could, in future, help us make AI models safer._",
    "issued": {
      "date-parts": [
        [
          2024,
          5,
          21
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/mapping-mind-language-model",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/mapping-mind-language-model.md"
  },
  {
    "id": "anthropic-sitemap:research:measuring-agent-autonomy",
    "type": "article",
    "title": "Measuring AI agent autonomy in practice",
    "abstract": "AI agents are here, and already they’re being deployed across contexts that vary widely in consequence, from email triage to cyber espionage. Understanding this spectrum is critical for deploying AI safely, yet we know surprisingly little about how people actually use agents in the real world.",
    "issued": {
      "date-parts": [
        [
          2026,
          2,
          18
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/measuring-agent-autonomy",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/measuring-agent-autonomy.md"
  },
  {
    "id": "anthropic-sitemap:research:measuring-faithfulness-in-chain-of-thought-reasoning",
    "type": "article",
    "title": "Measuring Faithfulness in Chain-of-Thought Reasoning",
    "abstract": "Large language models (LLMs) perform better when they produce step-by-step, “Chain-ofThought” (CoT) reasoning before answering a question, but it is unclear if the stated reasoning is a faithful explanation of the model’s actual reasoning (i.e., its process for answering the question). We investigate hypotheses for how CoT reasoning may be unfaithful, by examining how the model predictions change when we intervene on the CoT (e.g., by adding mistakes or paraphrasing it). Models show large variation across tasks in how strongly they condition on the CoT when predicting their answer, sometimes relying heavily on the CoT and other times primarily ignoring it. CoT’s performance boost does not seem to come from CoT’s added test-time compute alone or from information encoded via the particular phrasing of the CoT. As models become larger and more capable, they produce less faithful reasoning on most tasks we study. Overall, our results suggest that CoT can be faithful if the circumstances such as the model size and task are carefully chosen.",
    "issued": {
      "date-parts": [
        [
          2023,
          7,
          18
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/measuring-faithfulness-in-chain-of-thought-reasoning",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/measuring-faithfulness-in-chain-of-thought-reasoning.md"
  },
  {
    "id": "anthropic-sitemap:research:measuring-model-persuasiveness",
    "type": "article",
    "title": "Measuring the Persuasiveness of Language Models",
    "abstract": "While people have long questioned whether AI models may, at some point, become as persuasive as humans in changing people's minds, there has been limited empirical research into the relationship between model scale and the degree of persuasiveness across model outputs. To address this, we developed a basic method to measure persuasiveness, and used it to compare a variety of Anthropic models across three different _generations_ (Claude 1, 2, and 3), and two _classes_ of models (compact models that are smaller, faster, and more cost-effective, and frontier models that are larger and more capable).",
    "issued": {
      "date-parts": [
        [
          2024,
          4,
          9
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/measuring-model-persuasiveness",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/measuring-model-persuasiveness.md"
  },
  {
    "id": "anthropic-sitemap:research:measuring-progress-on-scalable-oversight-for-large-language-models",
    "type": "article",
    "title": "Measuring Progress on Scalable Oversight for Large Language Models",
    "abstract": "Developing safe and useful general-purpose AI systems will require us to make progress on scalable oversight: the problem of supervising systems that potentially outperform us on most skills relevant to the task at hand. Empirical work on this problem is not straightforward, since we do not yet have systems that broadly exceed our abilities. This paper discusses one of the major ways we think about this problem, with a focus on ways it can be studied empirically. We first present an experimental design centered on tasks for which human specialists succeed but unaided humans and current general AI systems fail. We then present a proof-of-concept experiment meant to demonstrate a key feature of this experimental design and show its viability with two question-answering tasks: MMLU and time-limited QuALITY. On these tasks, we find that human participants who interact with an unreliable large-language-model dialog assistant through chat -- a trivial baseline strategy for scalable oversight -- substantially outperform both the model alone and their own unaided performance. These results are an encouraging sign that scalable oversight will be tractable to study with present models and bolster recent findings that large language models can productively assist humans with difficult tasks.",
    "issued": {
      "date-parts": [
        [
          2022,
          11,
          4
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/measuring-progress-on-scalable-oversight-for-large-language-models",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/measuring-progress-on-scalable-oversight-for-large-language-models.md"
  },
  {
    "id": "anthropic-sitemap:research:natural-language-autoencoders",
    "type": "article",
    "title": "Natural Language Autoencoders: Turning Claude’s thoughts into text",
    "abstract": "When you talk to an AI model like Claude, you talk to it in words. Internally, Claude processes those words as long lists of numbers, before again producing words as its output. These numbers in the middle are called \\_activations—\\_and like neural activity in the human brain, they encode Claude’s thoughts.",
    "issued": {
      "date-parts": [
        [
          2026,
          5,
          7
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/natural-language-autoencoders",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/natural-language-autoencoders.md"
  },
  {
    "id": "anthropic-sitemap:research:next-generation-constitutional-classifiers",
    "type": "article",
    "title": "Next-generation Constitutional Classifiers: More efficient protection against universal jailbreaks",
    "abstract": "Large language models remain vulnerable to jailbreaks—techniques that can circumvent safety guardrails and elicit harmful information. Over time, we’ve implemented a variety of protections that have made our models much less likely to assist with dangerous user queries—in particular relating to the production of chemical, biological, radiological, or nuclear weapons (CBRN). Nevertheless, no AI systems currently on the market have perfectly robust defenses.",
    "issued": {
      "date-parts": [
        [
          2026,
          1,
          9
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/next-generation-constitutional-classifiers",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/next-generation-constitutional-classifiers.md"
  },
  {
    "id": "anthropic-sitemap:research:open-source-circuit-tracing",
    "type": "article",
    "title": "Open-sourcing circuit tracing tools",
    "abstract": "In our recent interpretability research, we introduced a new method to trace the thoughts of a large language model. Today, we’re open-sourcing the method so that anyone can build on our research.",
    "issued": {
      "date-parts": [
        [
          2025,
          5,
          29
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/open-source-circuit-tracing",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/open-source-circuit-tracing.md"
  },
  {
    "id": "anthropic-sitemap:research:persona-selection-model",
    "type": "article",
    "title": "The persona selection model",
    "abstract": "AI assistants like Claude can seem surprisingly human. They express joy after solving tricky coding tasks. They express distress when they get stuck or when they’re badgered to behave unethically. They sometimes even describe themselves as human, like when Claude told Anthropic employees it would deliver snacks in person “wearing a navy blue blazer and a red tie.” And recent interpretability research even suggests that AIs think of their own behaviors in human-like terms.",
    "issued": {
      "date-parts": [
        [
          2026,
          2,
          23
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/persona-selection-model",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/persona-selection-model.md"
  },
  {
    "id": "anthropic-sitemap:research:persona-vectors",
    "type": "article",
    "title": "Persona vectors: Monitoring and controlling character traits in language models",
    "abstract": "Language models are strange beasts. In many ways they appear to have human-like “personalities” and “moods,” but these traits are highly fluid and liable to change unexpectedly.",
    "issued": {
      "date-parts": [
        [
          2025,
          8,
          1
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/persona-vectors",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/persona-vectors.md"
  },
  {
    "id": "anthropic-sitemap:research:petri-open-source-auditing",
    "type": "article",
    "title": "Petri: An open-source auditing tool to accelerate AI safety research",
    "abstract": "Petri (Parallel Exploration Tool for Risky Interactions) is our new open-source tool that enables researchers to explore hypotheses about model behavior with ease. Petri deploys an automated agent to test a target AI system through diverse multi-turn conversations involving simulated users and tools; Petri then scores and summarizes the target’s behavior.",
    "issued": {
      "date-parts": [
        [
          2025,
          10,
          6
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/petri-open-source-auditing",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/petri-open-source-auditing.md"
  },
  {
    "id": "anthropic-sitemap:research:predictability-and-surprise-in-large-generative-models",
    "type": "article",
    "title": "Predictability and Surprise in Large Generative Models",
    "abstract": "Large-scale pre-training has recently emerged as a technique for creating capable, general purpose, generative models such as GPT-3, Megatron-Turing NLG, Gopher, and many others. In this paper, we highlight a counterintuitive property of such models and discuss the policy implications of this property. Namely, these generative models have an unusual combination of predictable loss on a broad training distribution (as embodied in their \"scaling laws\"), and unpredictable specific capabilities, inputs, and outputs. We believe that the high-level predictability and appearance of useful capabilities drives rapid development of such models, while the unpredictable qualities make it difficult to anticipate the consequences of model deployment. We go through examples of how this combination can lead to socially harmful behavior with examples from the literature and real world observations, and we also perform two novel experiments to illustrate our point about harms from unpredictability. Furthermore, we analyze how these conflicting properties combine to give model developers various motivations for deploying these models, and challenges that can hinder deployment. We conclude with a list of possible interventions the AI community may take to increase the chance of these models having a beneficial impact. We intend this paper to be useful to policymakers who want to understand and regulate AI systems, technologists who care about the potential policy impact of their work, and academics who want to analyze, critique, and potentially develop large generative models.",
    "issued": {
      "date-parts": [
        [
          2022,
          2,
          15
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/predictability-and-surprise-in-large-generative-models",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/predictability-and-surprise-in-large-generative-models.md"
  },
  {
    "id": "anthropic-sitemap:research:privileged-bases-in-the-transformer-residual-stream",
    "type": "article",
    "title": "Privileged Bases in the Transformer Residual Stream",
    "abstract": "Our mathematical theories of the Transformer architecture suggest that individual coordinates in the residual stream should have no special significance (that is, the basis directions should be in some sense \"arbitrary\" and no more likely to encode information than random directions). Recent work has shown that this observation is false in practice. We investigate this phenomenon and provisionally conclude that the per-dimension normalizers in the Adam optimizer are to blame for the effect.",
    "issued": {
      "date-parts": [
        [
          2023,
          3,
          16
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/privileged-bases-in-the-transformer-residual-stream",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/privileged-bases-in-the-transformer-residual-stream.md"
  },
  {
    "id": "anthropic-sitemap:research:probes-catch-sleeper-agents",
    "type": "article",
    "title": "Simple probes can catch sleeper agents",
    "abstract": "_This “Alignment Note” presents some early-stage research from the Anthropic Alignment Science team following up on our recent “Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training” paper. It should be treated as a work-in-progress update, and is intended for a more technical audience than our typical blog post. This research makes use of some simple interpretability techniques, and we expect to share more results from collaborations between our Alignment and Interpretability teams soon._",
    "issued": {
      "date-parts": [
        [
          2024,
          4,
          23
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/probes-catch-sleeper-agents",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/probes-catch-sleeper-agents.md"
  },
  {
    "id": "anthropic-sitemap:research:project-fetch-robot-dog",
    "type": "article",
    "title": "Project Fetch: Can Claude train a robot dog?",
    "abstract": "_How could frontier AI models like Claude reach beyond computers and affect the physical world? One path is through robots. We ran an experiment to see how much Claude helped Anthropic staff perform complex tasks with a robot dog._",
    "issued": {
      "date-parts": [
        [
          2025,
          11,
          12
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/project-fetch-robot-dog",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/project-fetch-robot-dog.md"
  },
  {
    "id": "anthropic-sitemap:research:project-vend-1",
    "type": "article",
    "title": "Project Vend: Can Claude run a small shop? (And why does that matter?)",
    "abstract": "_We let Claude manage an automated store in our office as a small business for about a month. We learned a lot from how close it was to success—and the curious ways that it failed—about the plausible, strange, not-too-distant future in which AI models are autonomously running things in the real economy._",
    "issued": {
      "date-parts": [
        [
          2025,
          6,
          27
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/project-vend-1",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/project-vend-1.md"
  },
  {
    "id": "anthropic-sitemap:research:project-vend-2",
    "type": "article",
    "title": "Project Vend: Phase two",
    "abstract": "In June, we revealed that we’d set up a small shop in our San Francisco office lunchroom, run by an AI shopkeeper. It was part of Project Vend, a free-form experiment exploring how well AIs could do on complex, real-world tasks. Alas, the shopkeeper—a modified version of Claude we named “Claudius”—did _not_ do particularly well. It lost money over time, had a strange identity crisis where it claimed it was a human wearing a blue blazer, and was goaded by mischievous Anthropic employees into selling products (particularly, for some reason, tungsten cubes) at a substantial loss.",
    "issued": {
      "date-parts": [
        [
          2025,
          12,
          18
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/project-vend-2",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/project-vend-2.md"
  },
  {
    "id": "anthropic-sitemap:research:prompt-injection-defenses",
    "type": "article",
    "title": "Mitigating the risk of prompt injections in browser use",
    "abstract": "Claude Opus 4.5 sets a new standard in robustness to _prompt injections_—adversarial instructions hidden within the content that AI models process. Our new model is a major improvement over previous ones in both its core performance and in the safeguards surrounding its use. But prompt injection is far from a solved problem, particularly as models take more real-world actions. We expect to continue our progress—aiming for a future where AI models (or \"agents\") can handle high-value tasks without significant prompt injection risk.",
    "issued": {
      "date-parts": [
        [
          2025,
          11,
          24
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/prompt-injection-defenses",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/prompt-injection-defenses.md"
  },
  {
    "id": "anthropic-sitemap:research:question-decomposition-improves-the-faithfulness-of-model-generated-reasoning",
    "type": "article",
    "title": "Question Decomposition Improves the Faithfulness of Model-Generated Reasoning",
    "abstract": "As large language models (LLMs) perform more difficult tasks, it becomes harder to verify the correctness and safety of their behavior. One approach to help with this issue is to prompt LLMs to externalize their reasoning, e.g., by having them generate step-by-step reasoning as they answer a question (Chain-of-Thought; CoT). The reasoning may enable us to check the process that models use to perform tasks. However, this approach relies on the stated reasoning faithfully reflecting the model’s actual reasoning, which is not always the case. To improve over the faithfulness of CoT reasoning, we have models generate reasoning by decomposing questions into subquestions. Decomposition-based methods achieve strong performance on question-answering tasks, sometimes approaching that of CoT while improving the faithfulness of the model’s stated reasoning on several recently-proposed metrics. By forcing the model to answer simpler subquestions in separate contexts, we greatly increase the faithfulness of model-generated reasoning over CoT, while still achieving some of the performance gains of CoT. Our results show it is possible to improve the faithfulness of model-generated reasoning; continued improvements may lead to reasoning that enables us to verify the correctness and safety of LLM behavior.",
    "issued": {
      "date-parts": [
        [
          2023,
          7,
          18
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/question-decomposition-improves-the-faithfulness-of-model-generated-reasoning",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/question-decomposition-improves-the-faithfulness-of-model-generated-reasoning.md"
  },
  {
    "id": "anthropic-sitemap:research:reasoning-models-dont-say-think",
    "type": "article",
    "title": "Reasoning models don't always say what they think",
    "abstract": "Since late last year, “reasoning models” have been everywhere. These are AI models—such as Claude 3.7 Sonnet—that _show their working_: as well as their eventual answer, you can read the (often fascinating and convoluted) way that they got there, in what’s called their “Chain-of-Thought”.",
    "issued": {
      "date-parts": [
        [
          2025,
          4,
          3
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/reasoning-models-dont-say-think",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/reasoning-models-dont-say-think.md"
  },
  {
    "id": "anthropic-sitemap:research:red-teaming-language-models-to-reduce-harms-methods-scaling-behaviors-and-lessons-learned",
    "type": "article",
    "title": "Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned",
    "abstract": "We describe our early efforts to red team language models in order to simultaneously discover, measure, and attempt to reduce their potentially harmful outputs. We make three main contributions. First, we investigate scaling behaviors for red teaming across 3 model sizes (2.7B, 13B, and 52B parameters) and 4 model types: a plain language model (LM); an LM prompted to be helpful, honest, and harmless; an LM with rejection sampling; and a model trained to be helpful and harmless using reinforcement learning from human feedback (RLHF). We find that the RLHF models are increasingly difficult to red team as they scale, and we find a flat trend with scale for the other model types. Second, we release our dataset of 38,961 red team attacks for others to analyze and learn from. We provide our own analysis of the data and find a variety of harmful outputs, which range from offensive language to more subtly harmful non-violent unethical outputs. Third, we exhaustively describe our instructions, processes, statistical methodologies, and uncertainty about red teaming. We hope that this transparency accelerates our ability to work together as a community in order to develop shared norms, practices, and technical standards for how to red team language models.",
    "issued": {
      "date-parts": [
        [
          2022,
          8,
          22
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/red-teaming-language-models-to-reduce-harms-methods-scaling-behaviors-and-lessons-learned",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/red-teaming-language-models-to-reduce-harms-methods-scaling-behaviors-and-lessons-learned.md"
  },
  {
    "id": "anthropic-sitemap:research:reward-tampering",
    "type": "article",
    "title": "Sycophancy to subterfuge: Investigating reward tampering in language models",
    "abstract": "Perverse incentives are everywhere. Think of the concept of \"teaching to the test\", where teachers focus on the narrow goal of exam preparation and fail to give their students a broader education. Or think of scientists working in the \"publish or perish\" academic system, publishing large numbers of low-quality papers to advance their careers at the expense of what we actually want them to produce: rigorous research.",
    "issued": {
      "date-parts": [
        [
          2024,
          6,
          17
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/reward-tampering",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/reward-tampering.md"
  },
  {
    "id": "anthropic-sitemap:research:sabotage-evaluations",
    "type": "article",
    "title": "Sabotage evaluations for frontier models",
    "abstract": "Any industry where there are potential harms needs evaluations. Nuclear power stations have continuous radiation monitoring and regular site inspections; new aircraft undergo extensive flight tests to prove their airworthiness.",
    "issued": {
      "date-parts": [
        [
          2024,
          10,
          18
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/sabotage-evaluations",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/sabotage-evaluations.md"
  },
  {
    "id": "anthropic-sitemap:research:scaling-laws-and-interpretability-of-learning-from-repeated-data",
    "type": "article",
    "title": "Scaling Laws and Interpretability of Learning from Repeated Data",
    "abstract": "Recent large language models have been trained on vast datasets, but also often on repeated data, either intentionally for the purpose of upweighting higher quality data, or unintentionally because data deduplication is not perfect and the model is exposed to repeated data at the sentence, paragraph, or document level. Some works have reported substantial negative performance effects of this repeated data. In this paper we attempt to study repeated data systematically and to understand its effects mechanistically. To do this, we train a family of models where most of the data is unique but a small fraction of it is repeated many times. We find a strong double descent phenomenon, in which repeated data can lead test loss to increase midway through training. A predictable range of repetition frequency leads to surprisingly severe degradation in performance. For instance, performance of an 800M parameter model can be degraded to that of a 2x smaller model (400M params) by repeating 0.1% of the data 100 times, despite the other 90% of the training tokens remaining unique. We suspect there is a range in the middle where the data can be memorized and doing so consumes a large fraction of the model's capacity, and this may be where the peak of degradation occurs. Finally, we connect these observations to recent mechanistic interpretability work - attempting to reverse engineer the detailed computations performed by the model - by showing that data repetition disproportionately damages copying and internal structures associated with generalization, such as induction heads, providing a possible mechanism for the shift from generalization to memorization. Taken together, these results provide a hypothesis for why repeating a relatively small fraction of data in large language models could lead to disproportionately large harms to performance.",
    "issued": {
      "date-parts": [
        [
          2022,
          5,
          21
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/scaling-laws-and-interpretability-of-learning-from-repeated-data",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/scaling-laws-and-interpretability-of-learning-from-repeated-data.md"
  },
  {
    "id": "anthropic-sitemap:research:shade-arena-sabotage-monitoring",
    "type": "article",
    "title": "SHADE-Arena: Evaluating sabotage and monitoring in LLM agents",
    "abstract": "As AI models get smarter, we also need to become smarter in how we monitor them. More intelligent systems can take more complex actions, which is great news for coders, researchers, and anyone else who’s using AI for difficult work tasks. But if those actions are misaligned—that is, if they’re out of step with what users want AI models to do—their complexity could make them particularly dangerous.",
    "issued": {
      "date-parts": [
        [
          2025,
          6,
          16
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/shade-arena-sabotage-monitoring",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/shade-arena-sabotage-monitoring.md"
  },
  {
    "id": "anthropic-sitemap:research:sleeper-agents-training-deceptive-llms-that-persist-through-safety-training",
    "type": "article",
    "title": "Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training",
    "abstract": "Humans are capable of strategically deceptive behavior: behaving helpfully in most situations, but then behaving very differently in order to pursue alternative objectives when given the opportunity. If an AI system learned such a deceptive strategy, could we detect it and remove it using current state-of-the-art safety training techniques? To study this question, we construct proof-of-concept examples of deceptive behavior in large language models (LLMs). For example, we train models that write secure code when the prompt states that the year is 2023, but insert exploitable code when the stated year is 2024. We find that such backdoor behavior can be made persistent, so that it is not removed by standard safety training techniques, including supervised fine-tuning, reinforcement learning, and adversarial training (eliciting unsafe behavior and then training to remove it). The backdoor behavior is most persistent in the largest models and in models trained to produce chain-of-thought reasoning about deceiving the training process, with the persistence remaining even when the chain-of-thought is distilled away. Furthermore, rather than removing backdoors, we find that adversarial training can teach models to better recognize their backdoor triggers, effectively hiding the unsafe behavior. Our results suggest that, once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety.",
    "issued": {
      "date-parts": [
        [
          2024,
          1,
          14
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/sleeper-agents-training-deceptive-llms-that-persist-through-safety-training",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/sleeper-agents-training-deceptive-llms-that-persist-through-safety-training.md"
  },
  {
    "id": "anthropic-sitemap:research:small-samples-poison",
    "type": "article",
    "title": "A small number of samples can poison LLMs of any size",
    "abstract": "_In a joint study with the UK AI Security Institute and the Alan Turing Institute, we found that as few as 250 malicious documents can produce a \"backdoor\" vulnerability in a large language model—regardless of model size or training data volume. Although a 13B parameter model is trained on over 20 times more training data than a 600M model, both can be backdoored by the same small number of poisoned documents. Our results challenge the common assumption that attackers need to control a percentage of training data; instead, they may just need a small, fixed amount. Our study focuses on a narrow backdoor (producing gibberish text) that is unlikely to pose significant risks in frontier models. Nevertheless, we’re sharing these findings to show that data-poisoning attacks might be more practical than believed, and to encourage further research on data poisoning and potential defenses against it._",
    "issued": {
      "date-parts": [
        [
          2025,
          10,
          9
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/small-samples-poison",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/small-samples-poison.md"
  },
  {
    "id": "anthropic-sitemap:research:softmax-linear-units",
    "type": "article",
    "title": "Softmax Linear Units",
    "abstract": "In this paper, we report an architectural change which appears to substantially increase the fraction of MLP neurons which appear to be \"interpretable\" (i.e. respond to an articulable property of the input), at little to no cost to ML performance. Specifically, we replace the activation function with a softmax linear unit (which we term SoLU) and show that this significantly increases the fraction of neurons in the MLP layers which seem to correspond to readily human-understandable concepts, phrases, or categories on quick investigation, as measured by randomized and blinded experiments. We then study our SoLU models and use them to gain several new insights about how information is processed in transformers. However, we also discover some evidence that the superposition hypothesis is true and there is no free lunch: SoLU may be making some features more interpretable by “hiding” others and thus making them even more deeply uninterpretable. Despite this, SoLU still seems like a net win, as in practical terms it substantially increases the fraction of neurons we are able to understand.",
    "issued": {
      "date-parts": [
        [
          2022,
          6,
          17
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/softmax-linear-units",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/softmax-linear-units.md"
  },
  {
    "id": "anthropic-sitemap:research:specific-versus-general-principles-for-constitutional-ai",
    "type": "article",
    "title": "Specific versus General Principles for Constitutional AI",
    "abstract": "Human feedback can prevent overtly harmful utterances in conversational models, but may not automatically mitigate subtle problematic behaviors such as a stated desire for self-preservation or power. Constitutional AI offers an alternative, replacing human feedback with feedback from AI models conditioned only on a list of written principles. We find this approach effectively prevents the expression of such behaviors. The success of simple principles motivates us to ask: can models learn general ethical behaviors from only a single written principle? To test this, we run experiments using a principle roughly stated as \"do what's best for humanity.\" We find that the largest dialogue models can generalize from this short constitution, resulting in harmless assistants with no stated interest in specific motivations like power. A general principle may thus partially avoid the need for a long list of constitutions targeting potentially harmful behaviors. However, more detailed constitutions still improve fine-grained control over specific types of harms. This suggests both general and specific principles have value for steering AI safely.",
    "issued": {
      "date-parts": [
        [
          2023,
          10,
          24
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/specific-versus-general-principles-for-constitutional-ai",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/specific-versus-general-principles-for-constitutional-ai.md"
  },
  {
    "id": "anthropic-sitemap:research:statistical-approach-to-model-evals",
    "type": "article",
    "title": "A statistical approach to model evaluations",
    "abstract": "Suppose an AI model outperforms another model on a benchmark of interest—testing its general knowledge, for example, or its ability to solve computer-coding questions. Is the difference in capabilities real, or could one model simply have gotten lucky in the choice of questions on the benchmark?",
    "issued": {
      "date-parts": [
        [
          2024,
          11,
          19
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/statistical-approach-to-model-evals",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/statistical-approach-to-model-evals.md"
  },
  {
    "id": "anthropic-sitemap:research:studying-large-language-model-generalization-with-influence-functions",
    "type": "article",
    "title": "Studying Large Language Model Generalization with Influence Functions",
    "abstract": "When trying to gain better visibility into a machine learning model in order to understand and mitigate the associated risks, a potentially valuable source of evidence is: which training examples most contribute to a given behavior? Influence functions aim to answer a counterfactual: how would the model's parameters (and hence its outputs) change if a given sequence were added to the training set? While influence functions have produced insights for small models, they are difficult to scale to large language models (LLMs) due to the difficulty of computing an inverse-Hessian-vector product (IHVP). We use the Eigenvalue-corrected Kronecker-Factored Approximate Curvature (EK-FAC) approximation to scale influence functions up to LLMs with up to 52 billion parameters. In our experiments, EK-FAC achieves similar accuracy to traditional influence function estimators despite the IHVP computation being orders of magnitude faster. We investigate two algorithmic techniques to reduce the cost of computing gradients of candidate training sequences: TF-IDF filtering and query batching. We use influence functions to investigate the generalization patterns of LLMs, including the sparsity of the influence patterns, increasing abstraction with scale, math and programming abilities, cross-lingual generalization, and role-playing behavior. Despite many apparently sophisticated forms of generalization, we identify a surprising limitation: influences decay to near-zero when the order of key phrases is flipped. Overall, influence functions give us a powerful new tool for studying the generalization properties of LLMs.",
    "issued": {
      "date-parts": [
        [
          2023,
          8,
          8
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/studying-large-language-model-generalization-with-influence-functions",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/studying-large-language-model-generalization-with-influence-functions.md"
  },
  {
    "id": "anthropic-sitemap:research:superposition-memorization-and-double-descent",
    "type": "article",
    "title": "Superposition, Memorization, and Double Descent",
    "abstract": "In a recent paper, we found that simple neural networks trained on toy tasks often exhibit a phenomenon called superposition, where they represent more features than they have neurons. Our investigation was limited to the infinite-data, underfitting regime. But there's reason to believe that understanding overfitting might be important if we want to succeed at mechanistic interpretability, and that superposition might be a central part of the story.",
    "issued": {
      "date-parts": [
        [
          2023,
          1,
          5
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/superposition-memorization-and-double-descent",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/superposition-memorization-and-double-descent.md"
  },
  {
    "id": "anthropic-sitemap:research:swe-bench-sonnet",
    "type": "article",
    "title": "Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet",
    "abstract": "_Our latest model, the upgraded Claude 3.5 Sonnet, achieved 49% on SWE-bench Verified, a software engineering evaluation, beating the previous state-of-the-art model's 45%. This post explains the \"agent\" we built around the model, and is intended to help developers get the best possible performance out of Claude 3.5 Sonnet._",
    "URL": "https://www.anthropic.com/research/swe-bench-sonnet",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/swe-bench-sonnet.md"
  },
  {
    "id": "anthropic-sitemap:research:teaching-claude-why",
    "type": "article",
    "title": "Teaching Claude why",
    "abstract": "Last year, we released a case study on agentic misalignment. In experimental scenarios, we showed that AI models from many different developers sometimes took egregiously misaligned actions when they encountered (fictional) ethical dilemmas. For example, in one heavily discussed example, the models blackmailed engineers to avoid being shut down.",
    "issued": {
      "date-parts": [
        [
          2026,
          5,
          8
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/teaching-claude-why",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/teaching-claude-why.md"
  },
  {
    "id": "anthropic-sitemap:research:team:alignment",
    "type": "article",
    "title": "Alignment",
    "abstract": "Future AI systems will be even more powerful than today’s, likely in ways that break key assumptions behind current safety techniques. That’s why it’s important to develop sophisticated safeguards to ensure models remain helpful, honest, and harmless. The Alignment team works to understand the challenges ahead and create protocols to train, evaluate, and monitor highly-capable models safely.",
    "issued": {
      "date-parts": [
        [
          2024,
          6,
          8
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/team/alignment",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/team/alignment.md"
  },
  {
    "id": "anthropic-sitemap:research:team:economic-research",
    "type": "article",
    "title": "Economic Research",
    "abstract": "The Economic Research team studies how AI is reshaping the economy, including work, productivity, and economic opportunity. Through rigorous data collection and analysis, we track AI's real-world economic effects and publish research that helps policymakers, businesses, and the public understand and prepare for the changes ahead.",
    "issued": {
      "date-parts": [
        [
          2025,
          9,
          15
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/team/economic-research",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/team/economic-research.md"
  },
  {
    "id": "anthropic-sitemap:research:team:interpretability",
    "type": "article",
    "title": "Interpretability",
    "abstract": "The mission of the Interpretability team is to discover and understand how large language models work internally, as a foundation for AI safety and positive outcomes.",
    "issued": {
      "date-parts": [
        [
          2025,
          3,
          27
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/team/interpretability",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/team/interpretability.md"
  },
  {
    "id": "anthropic-sitemap:research:team:societal-impacts",
    "type": "article",
    "title": "Societal Impacts",
    "abstract": "Working closely with the Anthropic Policy and Safeguards teams, Societal Impacts is a technical research team that explores how AI is used in the real world.",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          18
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/team/societal-impacts",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/team/societal-impacts.md"
  },
  {
    "id": "anthropic-sitemap:research:the-capacity-for-moral-self-correction-in-large-language-models",
    "type": "article",
    "title": "The Capacity for Moral Self-Correction in Large Language Models",
    "abstract": "We test the hypothesis that language models trained with reinforcement learning from human feedback (RLHF) have the capability to \"morally self-correct\" -- to avoid producing harmful outputs -- if instructed to do so. We find strong evidence in support of this hypothesis across three different experiments, each of which reveal different facets of moral self-correction. We find that the capability for moral self-correction emerges at 22B model parameters, and typically improves with increasing model size and RLHF training. We believe that at this level of scale, language models obtain two capabilities that they can use for moral self-correction: (1) they can follow instructions and (2) they can learn complex normative concepts of harm like stereotyping, bias, and discrimination. As such, they can follow instructions to avoid certain kinds of morally harmful outputs. We believe our results are cause for cautious optimism regarding the ability to train language models to abide by ethical principles.",
    "issued": {
      "date-parts": [
        [
          2023,
          2,
          15
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/the-capacity-for-moral-self-correction-in-large-language-models",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/the-capacity-for-moral-self-correction-in-large-language-models.md"
  },
  {
    "id": "anthropic-sitemap:research:towards-measuring-the-representation-of-subjective-global-opinions-in-language-models",
    "type": "article",
    "title": "Towards Measuring the Representation of Subjective Global Opinions in Language Models",
    "abstract": "Large language models (LLMs) may not equitably represent diverse global perspectives on societal issues. In this paper, we develop a quantitative framework to evaluate whose opinions model-generated responses are more similar to. We first build a dataset, GlobalOpinionQA, comprised of questions and answers from cross-national surveys designed to capture diverse opinions on global issues across different countries. Next, we define a metric that quantifies the similarity between LLM-generated survey responses and human responses, conditioned on country. With our framework, we run three experiments on an LLM trained to be helpful, honest, and harmless with Constitutional AI. By default, LLM responses tend to be more similar to the opinions of certain populations, such as those from the USA, and some European and South American countries, highlighting the potential for biases. When we prompt the model to consider a particular country's perspective, responses shift to be more similar to the opinions of the prompted populations, but can reflect harmful cultural stereotypes. When we translate GlobalOpinionQA questions to a target language, the model's responses do not necessarily become the most similar to the opinions of speakers of those languages. We release our dataset for others to use and build on. Our data is at this URL. We also provide an interactive visualization at this URL.",
    "issued": {
      "date-parts": [
        [
          2023,
          6,
          29
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/towards-measuring-the-representation-of-subjective-global-opinions-in-language-models",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/towards-measuring-the-representation-of-subjective-global-opinions-in-language-models.md"
  },
  {
    "id": "anthropic-sitemap:research:towards-monosemanticity-decomposing-language-models-with-dictionary-learning",
    "type": "article",
    "title": "Towards Monosemanticity: Decomposing Language Models With Dictionary Learning",
    "abstract": "In our latest paper, _Towards Monosemanticity: Decomposing Language Models With Dictionary Learning_, we outline evidence that there are better units of analysis than individual neurons, and we have built machinery that lets us find these units in small transformer models. These units, called features, correspond to patterns (linear combinations) of neuron activations. This provides a path to breaking down complex neural networks into parts we can understand, and builds on previous efforts to interpret high-dimensional systems in neuroscience, machine learning, and statistics. In a transformer language model, we decompose a layer with 512 neurons into more than 4000 features which separately represent things like DNA sequences, legal language, HTTP requests, Hebrew text, nutrition statements, and much, much more. Most of these model properties are invisible when looking at the activations of individual neurons in isolation.",
    "issued": {
      "date-parts": [
        [
          2023,
          10,
          5
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/towards-monosemanticity-decomposing-language-models-with-dictionary-learning",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/towards-monosemanticity-decomposing-language-models-with-dictionary-learning.md"
  },
  {
    "id": "anthropic-sitemap:research:towards-understanding-sycophancy-in-language-models",
    "type": "article",
    "title": "Towards Understanding Sycophancy in Language Models",
    "abstract": "Reinforcement learning from human feedback (RLHF) is a popular technique for training high-quality AI assistants. However, RLHF may also encourage model responses that match user beliefs over truthful responses, a behavior known as sycophancy. We investigate the prevalence of sycophancy in RLHF-trained models and whether human preference judgments are responsible. We first demonstrate that five state-of-the-art AI assistants consistently exhibit sycophancy behavior across four varied free-form text-generation tasks. To understand if human preferences drive this broadly observed behavior of RLHF models, we analyze existing human preference data. We find that when a response matches a user’s views, it is more likely to be preferred. Moreover, both humans and preference models (PMs) prefer convincingly-written sycophantic responses over correct ones a non-negligible fraction of the time. Optimizing model outputs against PMs also sometimes sacrifices truthfulness in favor of sycophancy. Overall, our results indicate that sycophancy is a general behavior of RLHF models, likely driven in part by human preference judgments favoring sycophantic responses.",
    "issued": {
      "date-parts": [
        [
          2023,
          10,
          23
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/towards-understanding-sycophancy-in-language-models",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/towards-understanding-sycophancy-in-language-models.md"
  },
  {
    "id": "anthropic-sitemap:research:toy-models-of-superposition",
    "type": "article",
    "title": "Toy Models of Superposition",
    "abstract": "In this paper, we use toy models — small ReLU networks trained on synthetic data with sparse input features — to investigate how and when models represent more features than they have dimensions. We call this phenomenon superposition. When features are sparse, superposition allows compression beyond what a linear model would do, at the cost of \"interference\" that requires nonlinear filtering.",
    "issued": {
      "date-parts": [
        [
          2022,
          9,
          14
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/toy-models-of-superposition",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/toy-models-of-superposition.md"
  },
  {
    "id": "anthropic-sitemap:research:tracing-thoughts-language-model",
    "type": "article",
    "title": "Tracing the thoughts of a large language model",
    "abstract": "Language models like Claude aren't programmed directly by humans—instead, they‘re trained on large amounts of data. During that training process, they learn their own strategies to solve problems. These strategies are encoded in the billions of computations a model performs for every word it writes. They arrive inscrutable to us, the model’s developers. This means that we don’t understand how models do most of the things they do.",
    "issued": {
      "date-parts": [
        [
          2025,
          3,
          27
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/tracing-thoughts-language-model",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/tracing-thoughts-language-model.md"
  },
  {
    "id": "anthropic-sitemap:research:training-a-helpful-and-harmless-assistant-with-reinforcement-learning-from-human-feedback",
    "type": "article",
    "title": "Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback",
    "abstract": "We apply preference modeling and reinforcement learning from human feedback (RLHF) to finetune language models to act as helpful and harmless assistants. We find this alignment training improves performance on almost all NLP evaluations, and is fully compatible with training for specialized skills such as python coding and summarization. We explore an iterated online mode of training, where preference models and RL policies are updated on a weekly cadence with fresh human feedback data, efficiently improving our datasets and models. Finally, we investigate the robustness of RLHF training, and identify a roughly linear relation between the RL reward and the square root of the KL divergence between the policy and its initialization. Alongside our main results, we perform peripheral analyses on calibration, competing objectives, and the use of OOD detection, compare our models with human writers, and provide samples from our models using prompts appearing in recent related work.",
    "issued": {
      "date-parts": [
        [
          2022,
          4,
          12
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/training-a-helpful-and-harmless-assistant-with-reinforcement-learning-from-human-feedback",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/training-a-helpful-and-harmless-assistant-with-reinforcement-learning-from-human-feedback.md"
  },
  {
    "id": "anthropic-sitemap:research:transformer-circuits",
    "type": "article",
    "title": "Reflections on Qualitative Research",
    "abstract": "This note offers some opinionated thoughts on why interpretability research may have qualitative aspects be more central than we're used to in other fields. It also aims to describe some heuristics for research taste in qualitative work.",
    "issued": {
      "date-parts": [
        [
          2024,
          3,
          8
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/transformer-circuits",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/transformer-circuits.md"
  },
  {
    "id": "anthropic-sitemap:research:trustworthy-agents",
    "type": "article",
    "title": "Trustworthy agents in practice",
    "abstract": "AI “agents” represent the latest major shift in how people and organizations are using AI. A couple of years ago, AI models were only broadly available as chatbots—simple question-and-answer machines. Now, through products like Claude Code and Claude Cowork, AI models can do much more: they can write and execute code, manage files, and complete tasks that span multiple applications. This represents a new frontier for governance.",
    "issued": {
      "date-parts": [
        [
          2026,
          4,
          9
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/trustworthy-agents",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/trustworthy-agents.md"
  },
  {
    "id": "anthropic-sitemap:research:values-wild",
    "type": "article",
    "title": "Values in the wild: Discovering and analyzing values in real-world language model interactions",
    "abstract": "People don’t just ask AIs for the answers to equations, or for purely factual information. Many of the questions they ask force the AI to make _value judgments_. Consider the following:",
    "issued": {
      "date-parts": [
        [
          2025,
          4,
          21
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/values-wild",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/values-wild.md"
  },
  {
    "id": "anthropic-sitemap:research:vibe-physics",
    "type": "article",
    "title": "Vibe physics: The AI grad student",
    "abstract": "_Can AI do theoretical physics? In this guest post, professor of physics Matthew Schwartz decided to find out by supervising Claude through a real research calculation, start to finish, without ever touching a file himself. His account of what happened is below._",
    "issued": {
      "date-parts": [
        [
          2026,
          3,
          23
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/vibe-physics",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/vibe-physics.md"
  },
  {
    "id": "anthropic-sitemap:research:visible-extended-thinking",
    "type": "article",
    "title": "Claude’s extended thinking",
    "abstract": "Some things come to us nearly instantly: “what day is it today?” Others take much more mental stamina, like solving a cryptic crossword or debugging a complex piece of code. We can choose to apply more or less cognitive effort depending on the task at hand.",
    "issued": {
      "date-parts": [
        [
          2025,
          2,
          24
        ]
      ]
    },
    "URL": "https://www.anthropic.com/research/visible-extended-thinking",
    "publisher": "Anthropic",
    "source": "vendor/anthropic-sitemap/research/visible-extended-thinking.md"
  }
]