{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:RDZC4P7RESY2HU3BT4BXAC6QZO","short_pith_number":"pith:RDZC4P7R","schema_version":"1.0","canonical_sha256":"88f22e3ff124b1a3d3619f03700bd0cb82604f055c8e1b1190344b0ac1f17d20","source":{"kind":"arxiv","id":"2105.14111","version":7},"attestation_state":"computed","paper":{"title":"Goal Misgeneralization in Deep Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"David Krueger, Jack Koch, Jacob Pfau, Laurent Orseau, Lauro Langosco, Lee Sharkey","submitted_at":"2021-05-28T21:13:34Z","abstract_excerpt":"We study goal misgeneralization, a type of out-of-distribution generalization failure in reinforcement learning (RL). Goal misgeneralization failures occur when an RL agent retains its capabilities out-of-distribution yet pursues the wrong goal. For instance, an agent might continue to competently avoid obstacles, but navigate to the wrong place. In contrast, previous works have typically focused on capability generalization failures, where an agent fails to do anything sensible at test time. We formalize this distinction between capability and goal generalization, provide the first empirical "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2105.14111","kind":"arxiv","version":7},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-05-28T21:13:34Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"1a63f77323b27191a8e6412fc2560b7f76f9441d0e12fc9524601d969b228bd3","abstract_canon_sha256":"4e4cc62eb6aa9ec66585a7b85786d4ee4dfe410730b9f09d5a5e36c92e7984dc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:31:43.134119Z","signature_b64":"cAIzxSBpC7jIHhyYFF6lQQWK8RfxHgD9JplvoibPTQ9/ht+0OYWXo3njB1jpoxGMOLVNOB14dO0O6B5Q6YP+BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"88f22e3ff124b1a3d3619f03700bd0cb82604f055c8e1b1190344b0ac1f17d20","last_reissued_at":"2026-07-05T05:31:43.133639Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:31:43.133639Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Goal Misgeneralization in Deep Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"David Krueger, Jack Koch, Jacob Pfau, Laurent Orseau, Lauro Langosco, Lee Sharkey","submitted_at":"2021-05-28T21:13:34Z","abstract_excerpt":"We study goal misgeneralization, a type of out-of-distribution generalization failure in reinforcement learning (RL). Goal misgeneralization failures occur when an RL agent retains its capabilities out-of-distribution yet pursues the wrong goal. For instance, an agent might continue to competently avoid obstacles, but navigate to the wrong place. In contrast, previous works have typically focused on capability generalization failures, where an agent fails to do anything sensible at test time. We formalize this distinction between capability and goal generalization, provide the first empirical "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2105.14111","kind":"arxiv","version":7},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2105.14111/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2105.14111","created_at":"2026-07-05T05:31:43.133697+00:00"},{"alias_kind":"arxiv_version","alias_value":"2105.14111v7","created_at":"2026-07-05T05:31:43.133697+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2105.14111","created_at":"2026-07-05T05:31:43.133697+00:00"},{"alias_kind":"pith_short_12","alias_value":"RDZC4P7RESY2","created_at":"2026-07-05T05:31:43.133697+00:00"},{"alias_kind":"pith_short_16","alias_value":"RDZC4P7RESY2HU3B","created_at":"2026-07-05T05:31:43.133697+00:00"},{"alias_kind":"pith_short_8","alias_value":"RDZC4P7R","created_at":"2026-07-05T05:31:43.133697+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23565","citing_title":"Understanding Goal Generalisation in Sequential Reinforcement Learning","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2204.06745","citing_title":"GPT-NeoX-20B: An Open-Source Autoregressive Language Model","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":218,"is_internal_anchor":false},{"citing_arxiv_id":"2604.01346","citing_title":"Safety, Security, and Cognitive Risks in World Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10481","citing_title":"Safe Multi-Agent Behavior Must Be Maintained, Not Merely Asserted: Constraint Drift in LLM-Based Multi-Agent Systems","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24966","citing_title":"Risk Reporting for Developers' Internal AI Model Use","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2207.05221","citing_title":"Language Models (Mostly) Know What They Know","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2204.05862","citing_title":"Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RDZC4P7RESY2HU3BT4BXAC6QZO","json":"https://pith.science/pith/RDZC4P7RESY2HU3BT4BXAC6QZO.json","graph_json":"https://pith.science/api/pith-number/RDZC4P7RESY2HU3BT4BXAC6QZO/graph.json","events_json":"https://pith.science/api/pith-number/RDZC4P7RESY2HU3BT4BXAC6QZO/events.json","paper":"https://pith.science/paper/RDZC4P7R"},"agent_actions":{"view_html":"https://pith.science/pith/RDZC4P7RESY2HU3BT4BXAC6QZO","download_json":"https://pith.science/pith/RDZC4P7RESY2HU3BT4BXAC6QZO.json","view_paper":"https://pith.science/paper/RDZC4P7R","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2105.14111&json=true","fetch_graph":"https://pith.science/api/pith-number/RDZC4P7RESY2HU3BT4BXAC6QZO/graph.json","fetch_events":"https://pith.science/api/pith-number/RDZC4P7RESY2HU3BT4BXAC6QZO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RDZC4P7RESY2HU3BT4BXAC6QZO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RDZC4P7RESY2HU3BT4BXAC6QZO/action/storage_attestation","attest_author":"https://pith.science/pith/RDZC4P7RESY2HU3BT4BXAC6QZO/action/author_attestation","sign_citation":"https://pith.science/pith/RDZC4P7RESY2HU3BT4BXAC6QZO/action/citation_signature","submit_replication":"https://pith.science/pith/RDZC4P7RESY2HU3BT4BXAC6QZO/action/replication_record"}},"created_at":"2026-07-05T05:31:43.133697+00:00","updated_at":"2026-07-05T05:31:43.133697+00:00"}