{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:A5SBQQMMO5JWGP7YIZCQ4V2AZF","short_pith_number":"pith:A5SBQQMM","schema_version":"1.0","canonical_sha256":"076418418c7753633ff846450e5740c9516e1df80644930bdc745699de5b1c0b","source":{"kind":"arxiv","id":"2504.09712","version":2},"attestation_state":"computed","paper":{"title":"The Structural Safety Generalization Problem","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.CR","authors_text":"Ethan Kosak-Hine, George Ingebretsen, Jason Zhang, Julius Broomfield, Kellin Pelrine, Reihaneh Iranmanesh, Reihaneh Rabbany, Sara Pieri, Tia Nasir, Tom Gibbs","submitted_at":"2025-04-13T20:21:08Z","abstract_excerpt":"LLM jailbreaks are a widespread safety challenge. Given this problem has not yet been tractable, we suggest targeting a key failure mechanism: the failure of safety to generalize across semantically equivalent inputs. We further focus the target by requiring desirable tractability properties of attacks to study: explainability, transferability between models, and transferability between goals. We perform red-teaming within this framework by uncovering new vulnerabilities to multi-turn, multi-image, and translation-based attacks. These attacks are semantically equivalent by our design to their "},"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":"2504.09712","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2025-04-13T20:21:08Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"5ce9371552e9b22956213a1149a670c7e114aa79c27f5c53983de2af56102a38","abstract_canon_sha256":"ffdaac23185ef7b6ecb9f91ae063f92b854da3aee9f636a76227e7ec92b32a6a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:13:24.131850Z","signature_b64":"UDs4UcebUk1DzsF7C+TGLX46YPxWxYJnQ4+8QnQBvag5canp9F2x8hUe7B2cgnQookGnwJeeli2x4dqDPBhfAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"076418418c7753633ff846450e5740c9516e1df80644930bdc745699de5b1c0b","last_reissued_at":"2026-07-05T11:13:24.131366Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:13:24.131366Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Structural Safety Generalization Problem","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.CR","authors_text":"Ethan Kosak-Hine, George Ingebretsen, Jason Zhang, Julius Broomfield, Kellin Pelrine, Reihaneh Iranmanesh, Reihaneh Rabbany, Sara Pieri, Tia Nasir, Tom Gibbs","submitted_at":"2025-04-13T20:21:08Z","abstract_excerpt":"LLM jailbreaks are a widespread safety challenge. Given this problem has not yet been tractable, we suggest targeting a key failure mechanism: the failure of safety to generalize across semantically equivalent inputs. We further focus the target by requiring desirable tractability properties of attacks to study: explainability, transferability between models, and transferability between goals. We perform red-teaming within this framework by uncovering new vulnerabilities to multi-turn, multi-image, and translation-based attacks. These attacks are semantically equivalent by our design to their "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.09712","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/2504.09712/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":"2504.09712","created_at":"2026-07-05T11:13:24.131426+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.09712v2","created_at":"2026-07-05T11:13:24.131426+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.09712","created_at":"2026-07-05T11:13:24.131426+00:00"},{"alias_kind":"pith_short_12","alias_value":"A5SBQQMMO5JW","created_at":"2026-07-05T11:13:24.131426+00:00"},{"alias_kind":"pith_short_16","alias_value":"A5SBQQMMO5JWGP7Y","created_at":"2026-07-05T11:13:24.131426+00:00"},{"alias_kind":"pith_short_8","alias_value":"A5SBQQMM","created_at":"2026-07-05T11:13:24.131426+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.18576","citing_title":"SafeWork-R1: Coevolving Safety and Intelligence under the AI-45$^{\\circ}$ Law","ref_index":3,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/A5SBQQMMO5JWGP7YIZCQ4V2AZF","json":"https://pith.science/pith/A5SBQQMMO5JWGP7YIZCQ4V2AZF.json","graph_json":"https://pith.science/api/pith-number/A5SBQQMMO5JWGP7YIZCQ4V2AZF/graph.json","events_json":"https://pith.science/api/pith-number/A5SBQQMMO5JWGP7YIZCQ4V2AZF/events.json","paper":"https://pith.science/paper/A5SBQQMM"},"agent_actions":{"view_html":"https://pith.science/pith/A5SBQQMMO5JWGP7YIZCQ4V2AZF","download_json":"https://pith.science/pith/A5SBQQMMO5JWGP7YIZCQ4V2AZF.json","view_paper":"https://pith.science/paper/A5SBQQMM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.09712&json=true","fetch_graph":"https://pith.science/api/pith-number/A5SBQQMMO5JWGP7YIZCQ4V2AZF/graph.json","fetch_events":"https://pith.science/api/pith-number/A5SBQQMMO5JWGP7YIZCQ4V2AZF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/A5SBQQMMO5JWGP7YIZCQ4V2AZF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/A5SBQQMMO5JWGP7YIZCQ4V2AZF/action/storage_attestation","attest_author":"https://pith.science/pith/A5SBQQMMO5JWGP7YIZCQ4V2AZF/action/author_attestation","sign_citation":"https://pith.science/pith/A5SBQQMMO5JWGP7YIZCQ4V2AZF/action/citation_signature","submit_replication":"https://pith.science/pith/A5SBQQMMO5JWGP7YIZCQ4V2AZF/action/replication_record"}},"created_at":"2026-07-05T11:13:24.131426+00:00","updated_at":"2026-07-05T11:13:24.131426+00:00"}