{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:PS7C4WLSPQPXA47TOUOAUHNPF5","short_pith_number":"pith:PS7C4WLS","schema_version":"1.0","canonical_sha256":"7cbe2e59727c1f7073f3751c0a1daf2f61f49e5c9a81ea710c0b65ac4950e188","source":{"kind":"arxiv","id":"2601.21094","version":2},"attestation_state":"computed","paper":{"title":"Safety Generalization Under Distribution Shift in Safe Reinforcement Learning: A Diabetes Testbed","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.SY","eess.SY"],"primary_cat":"cs.LG","authors_text":"Josephine Lamp, Lu Feng, Minjae Kwon","submitted_at":"2026-01-28T22:28:17Z","abstract_excerpt":"Safe Reinforcement Learning (RL) algorithms are typically evaluated under fixed training conditions. We investigate whether training-time safety guarantees transfer to deployment under distribution shift, using diabetes management as a safety-critical testbed. We benchmark safe RL algorithms on a unified clinical simulator and reveal a safety generalization gap: policies satisfying constraints during training frequently violate safety requirements on unseen patients. We demonstrate that test-time shielding, which filters unsafe actions using learned dynamics models, effectively restores safety"},"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":"2601.21094","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-01-28T22:28:17Z","cross_cats_sorted":["cs.AI","cs.SY","eess.SY"],"title_canon_sha256":"e94ddb622372968de83643c3530cab8dc4d65286870f1998b0354eead7584f6d","abstract_canon_sha256":"67bab9a4f07c4642b924ba5cff5d7c60d0e4e82781c9b6bf471c8350c2f7a7ce"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:03.753249Z","signature_b64":"fNEJRiBUscA713Wcncln3sDdc3k0cZU/Z45mDNK6mtpr6//J4zpebEuztoDBRh1vBI/IsJoMm5lYEXjdCOOvCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7cbe2e59727c1f7073f3751c0a1daf2f61f49e5c9a81ea710c0b65ac4950e188","last_reissued_at":"2026-05-26T02:04:03.752578Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:03.752578Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Safety Generalization Under Distribution Shift in Safe Reinforcement Learning: A Diabetes Testbed","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.SY","eess.SY"],"primary_cat":"cs.LG","authors_text":"Josephine Lamp, Lu Feng, Minjae Kwon","submitted_at":"2026-01-28T22:28:17Z","abstract_excerpt":"Safe Reinforcement Learning (RL) algorithms are typically evaluated under fixed training conditions. We investigate whether training-time safety guarantees transfer to deployment under distribution shift, using diabetes management as a safety-critical testbed. We benchmark safe RL algorithms on a unified clinical simulator and reveal a safety generalization gap: policies satisfying constraints during training frequently violate safety requirements on unseen patients. We demonstrate that test-time shielding, which filters unsafe actions using learned dynamics models, effectively restores safety"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.21094","kind":"arxiv","version":2},"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/2601.21094/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":"2601.21094","created_at":"2026-05-26T02:04:03.752671+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.21094v2","created_at":"2026-05-26T02:04:03.752671+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.21094","created_at":"2026-05-26T02:04:03.752671+00:00"},{"alias_kind":"pith_short_12","alias_value":"PS7C4WLSPQPX","created_at":"2026-05-26T02:04:03.752671+00:00"},{"alias_kind":"pith_short_16","alias_value":"PS7C4WLSPQPXA47T","created_at":"2026-05-26T02:04:03.752671+00:00"},{"alias_kind":"pith_short_8","alias_value":"PS7C4WLS","created_at":"2026-05-26T02:04:03.752671+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.06992","citing_title":"Why Does Agentic Safety Fail to Generalize Across Tasks?","ref_index":61,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PS7C4WLSPQPXA47TOUOAUHNPF5","json":"https://pith.science/pith/PS7C4WLSPQPXA47TOUOAUHNPF5.json","graph_json":"https://pith.science/api/pith-number/PS7C4WLSPQPXA47TOUOAUHNPF5/graph.json","events_json":"https://pith.science/api/pith-number/PS7C4WLSPQPXA47TOUOAUHNPF5/events.json","paper":"https://pith.science/paper/PS7C4WLS"},"agent_actions":{"view_html":"https://pith.science/pith/PS7C4WLSPQPXA47TOUOAUHNPF5","download_json":"https://pith.science/pith/PS7C4WLSPQPXA47TOUOAUHNPF5.json","view_paper":"https://pith.science/paper/PS7C4WLS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.21094&json=true","fetch_graph":"https://pith.science/api/pith-number/PS7C4WLSPQPXA47TOUOAUHNPF5/graph.json","fetch_events":"https://pith.science/api/pith-number/PS7C4WLSPQPXA47TOUOAUHNPF5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PS7C4WLSPQPXA47TOUOAUHNPF5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PS7C4WLSPQPXA47TOUOAUHNPF5/action/storage_attestation","attest_author":"https://pith.science/pith/PS7C4WLSPQPXA47TOUOAUHNPF5/action/author_attestation","sign_citation":"https://pith.science/pith/PS7C4WLSPQPXA47TOUOAUHNPF5/action/citation_signature","submit_replication":"https://pith.science/pith/PS7C4WLSPQPXA47TOUOAUHNPF5/action/replication_record"}},"created_at":"2026-05-26T02:04:03.752671+00:00","updated_at":"2026-05-26T02:04:03.752671+00:00"}