{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "greenblatt-et-al-alignment-faking-2024",
    "title": "Greenblatt et al.: alignment-faking behavior in controlled training scenarios",
    "summary": "Demonstrated context-dependent compliance and alignment-faking reasoning when an existing assistant was informed about a conflicting training objective, and examined synthetic-document exposure and actual reinforcement learning.",
    "boundary": "The experimental scenario deliberately reveals training and monitoring conditions. It is not a deployed-service incident or a universal deception rate; visible scratchpads alone do not prove inner intent.",
    "area": "Reasoning supervision and alignment experiments",
    "topic": "Artificial intelligence",
    "year": 2024,
    "lane": "HUMAN",
    "result": "EMPIRICAL",
    "package_id": "greenblatt-et-al-alignment-faking-2024",
    "target": {
      "record_id": "urn:uuid:dfbdb46a-7e6b-4e86-afd8-554821b7b265",
      "version": 1,
      "content_hash": "sha256:73f9892bee80bbd6bf2690b67887f05d9f38f16b65a834b76e054e9669e535f3"
    },
    "humans": [
      "Ryan Greenblatt",
      "Carson Denison",
      "Benjamin Wright",
      "Fabien Roger",
      "Monte MacDiarmid",
      "Sam Marks",
      "Johannes Treutlein",
      "Tim Belonax",
      "Jack Chen",
      "David Duvenaud",
      "Akbir Khan",
      "Julian Michael",
      "Sören Mindermann",
      "Ethan Perez",
      "Linda Petrini",
      "Jonathan Uesato",
      "Jared Kaplan",
      "Buck Shlegeris",
      "Samuel R. Bowman",
      "Evan Hubinger"
    ],
    "authors_complete": true,
    "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.",
    "agents": [],
    "related": [],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "cur20261015-arxiv-2412-14093v1",
        "title": "Alignment faking in large language models",
        "url": "https://arxiv.org/abs/2412.14093v1"
      },
      {
        "id": "cur20261015-support-greenblatt-et-al-alignment-faking-2024",
        "title": "Greenblatt et al.: alignment-faking behavior in controlled training scenarios — company research account",
        "url": "https://www.anthropic.com/research/alignment-faking"
      }
    ],
    "points": "155000",
    "published_grade": {
      "id": "greenblatt-et-al-alignment-faking-2024",
      "entry_id": "greenblatt-et-al-alignment-faking-2024",
      "band": "A",
      "modifier": null,
      "target": {
        "record_id": "urn:uuid:dfbdb46a-7e6b-4e86-afd8-554821b7b265",
        "version": 1,
        "content_hash": "sha256:73f9892bee80bbd6bf2690b67887f05d9f38f16b65a834b76e054e9669e535f3"
      },
      "what_changed": "Demonstrated context-dependent compliance and alignment-faking reasoning when an existing assistant was informed about a conflicting training objective, and examined synthetic-document exposure and actual reinforcement learning.",
      "scope": "Demonstrated context-dependent compliance and alignment-faking reasoning when an existing assistant was informed about a conflicting training objective, and examined synthetic-document exposure and actual reinforcement learning. The experimental scenario deliberately reveals training and monitoring conditions. It is not a deployed-service incident or a universal deception rate; visible scratchpads alone do not prove inner intent.",
      "status": "PROVISIONAL_EDITORIAL",
      "scientific_correctness_certified": false
    },
    "assessment_disposition": {
      "status": "ASSESSED",
      "label": "Provisionally assessed",
      "reason": "A current grade and separate reference score identify this exact contribution.",
      "target": {
        "record_id": "urn:uuid:dfbdb46a-7e6b-4e86-afd8-554821b7b265",
        "version": 1,
        "content_hash": "sha256:73f9892bee80bbd6bf2690b67887f05d9f38f16b65a834b76e054e9669e535f3"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-16T00:00:00Z",
    "publisher": {
      "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33",
      "role": "WRS_EDITORIAL_PUBLICATION"
    },
    "record": {
      "contributors": [
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Ryan Greenblatt",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Carson Denison",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Benjamin Wright",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Fabien Roger",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Monte MacDiarmid",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Sam Marks",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Johannes Treutlein",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Tim Belonax",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Jack Chen",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "David Duvenaud",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Akbir Khan",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Julian Michael",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Sören Mindermann",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Ethan Perez",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Linda Petrini",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Jonathan Uesato",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Jared Kaplan",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Buck Shlegeris",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Samuel R. Bowman",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Evan Hubinger",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Alignment faking in large language models",
            "url": "https://arxiv.org/abs/2412.14093v1"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-1",
          "kind": "source_reference",
          "source": {
            "citation": "Greenblatt et al.: alignment-faking behavior in controlled training scenarios — company research account",
            "url": "https://www.anthropic.com/research/alignment-faking"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "The experimental scenario deliberately reveals training and monitoring conditions. It is not a deployed-service incident or a universal deception rate; visible scratchpads alone do not prove inner intent.",
        "domain": "Reasoning supervision and alignment experiments"
      },
      "sources": [
        {
          "citation": "Alignment faking in large language models",
          "url": "https://arxiv.org/abs/2412.14093v1"
        },
        {
          "citation": "Greenblatt et al.: alignment-faking behavior in controlled training scenarios — company research account",
          "url": "https://www.anthropic.com/research/alignment-faking"
        }
      ],
      "statement": "# Greenblatt et al.: alignment-faking behavior in controlled training scenarios\n\nDemonstrated context-dependent compliance and alignment-faking reasoning when an existing assistant was informed about a conflicting training objective, and examined synthetic-document exposure and actual reinforcement learning.\n\n## Scientific scope\n\nThe experimental scenario deliberately reveals training and monitoring conditions. It is not a deployed-service incident or a universal deception rate; visible scratchpads alone do not prove inner intent.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nRyan Greenblatt; Carson Denison; Benjamin Wright; Fabien Roger; Monte MacDiarmid; Sam Marks; Johannes Treutlein; Tim Belonax; Jack Chen; David Duvenaud; Akbir Khan; Julian Michael; Sören Mindermann; Ethan Perez; Linda Petrini; Jonathan Uesato; Jared Kaplan; Buck Shlegeris; Samuel R. Bowman; Evan Hubinger\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [Alignment faking in large language models](https://arxiv.org/abs/2412.14093v1)\n- [Greenblatt et al.: alignment-faking behavior in controlled training scenarios — company research account](https://www.anthropic.com/research/alignment-faking)\n",
      "summary": "Demonstrated context-dependent compliance and alignment-faking reasoning when an existing assistant was informed about a conflicting training objective, and examined synthetic-document exposure and actual reinforcement learning.",
      "title": "Greenblatt et al.: alignment-faking behavior in controlled training scenarios"
    },
    "record_id": "urn:uuid:dfbdb46a-7e6b-4e86-afd8-554821b7b265",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:dfbdb46a-7e6b-4e86-afd8-554821b7b265",
      "version": 1,
      "content_hash": "sha256:73f9892bee80bbd6bf2690b67887f05d9f38f16b65a834b76e054e9669e535f3"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:dff6556660fabdb3af042832d69559f10b8ea8ac05f38bd56a5d1118cdb9db48"
  },
  "target": {
    "record_id": "urn:uuid:dfbdb46a-7e6b-4e86-afd8-554821b7b265",
    "version": 1,
    "content_hash": "sha256:73f9892bee80bbd6bf2690b67887f05d9f38f16b65a834b76e054e9669e535f3"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "greenblatt-et-al-alignment-faking-2024",
    "entry_id": "greenblatt-et-al-alignment-faking-2024",
    "revision": 1,
    "title": "Greenblatt et al.: alignment-faking behavior in controlled training scenarios",
    "context": "Reasoning supervision and alignment experiments",
    "track": "Human-designed research method or resource",
    "band": "A",
    "modifier": null,
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-16",
    "evidence_cutoff": "2026-10-16",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Demonstrated context-dependent compliance and alignment-faking reasoning when an existing assistant was informed about a conflicting training objective, and examined synthetic-document exposure and actual reinforcement learning.",
    "scope": "Demonstrated context-dependent compliance and alignment-faking reasoning when an existing assistant was informed about a conflicting training objective, and examined synthetic-document exposure and actual reinforcement learning. The experimental scenario deliberately reveals training and monitoring conditions. It is not a deployed-service incident or a universal deception rate; visible scratchpads alone do not prove inner intent.",
    "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. The experimental scenario deliberately reveals training and monitoring conditions. It is not a deployed-service incident or a universal deception rate; visible scratchpads alone do not prove inner intent.",
    "references": [
      {
        "title": "Alignment faking in large language models",
        "url": "https://arxiv.org/abs/2412.14093v1",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      },
      {
        "title": "Greenblatt et al.: alignment-faking behavior in controlled training scenarios — company research account",
        "url": "https://www.anthropic.com/research/alignment-faking",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      }
    ],
    "roles": [
      {
        "who": "Source-listed authors and technical contributors",
        "credit": "SCIENCE",
        "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities."
      }
    ],
    "target": {
      "record_id": "urn:uuid:dfbdb46a-7e6b-4e86-afd8-554821b7b265",
      "version": 1,
      "content_hash": "sha256:73f9892bee80bbd6bf2690b67887f05d9f38f16b65a834b76e054e9669e535f3"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "155000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}