{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:4EUURVWS5XN2ZIFZ3UTB66JPHK","short_pith_number":"pith:4EUURVWS","schema_version":"1.0","canonical_sha256":"e12948d6d2eddbaca0b9dd261f792f3ab20d04f9c128110212b98bcc8b7d703b","source":{"kind":"arxiv","id":"2502.14486","version":1},"attestation_state":"computed","paper":{"title":"How Jailbreak Defenses Work and Ensemble? A Mechanistic Investigation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CR","authors_text":"Shujun Liu, Siyuan Wang, Xuanjing Huang, Yuhang Lai, Zhongyu Wei, Zhuohang Long","submitted_at":"2025-02-20T12:07:40Z","abstract_excerpt":"Jailbreak attacks, where harmful prompts bypass generative models' built-in safety, raise serious concerns about model vulnerability. While many defense methods have been proposed, the trade-offs between safety and helpfulness, and their application to Large Vision-Language Models (LVLMs), are not well understood. This paper systematically examines jailbreak defenses by reframing the standard generation task as a binary classification problem to assess model refusal tendencies for both harmful and benign queries. We identify two key defense mechanisms: safety shift, which increases refusal rat"},"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":"2502.14486","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2025-02-20T12:07:40Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"46901bb89ef113bb0095eee0bfefc1cb8f58a58a681ecac9e2dd57ba1b0805de","abstract_canon_sha256":"6ae3d2f806e8310b808808e7cbf430dca23e9f052a0e3100f97272aff360b8dd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:17:27.393019Z","signature_b64":"KZJB3p6Li8vaaxT/INSBIb5Ndi0RBxeYEA2mpzj5Dfj34Z8h29pZEaVFpz4aba7YU1C0J+4DyDlwx70vryNrDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e12948d6d2eddbaca0b9dd261f792f3ab20d04f9c128110212b98bcc8b7d703b","last_reissued_at":"2026-07-05T10:17:27.392514Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:17:27.392514Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How Jailbreak Defenses Work and Ensemble? A Mechanistic Investigation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CR","authors_text":"Shujun Liu, Siyuan Wang, Xuanjing Huang, Yuhang Lai, Zhongyu Wei, Zhuohang Long","submitted_at":"2025-02-20T12:07:40Z","abstract_excerpt":"Jailbreak attacks, where harmful prompts bypass generative models' built-in safety, raise serious concerns about model vulnerability. While many defense methods have been proposed, the trade-offs between safety and helpfulness, and their application to Large Vision-Language Models (LVLMs), are not well understood. This paper systematically examines jailbreak defenses by reframing the standard generation task as a binary classification problem to assess model refusal tendencies for both harmful and benign queries. We identify two key defense mechanisms: safety shift, which increases refusal rat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.14486","kind":"arxiv","version":1},"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/2502.14486/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":"2502.14486","created_at":"2026-07-05T10:17:27.392562+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.14486v1","created_at":"2026-07-05T10:17:27.392562+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.14486","created_at":"2026-07-05T10:17:27.392562+00:00"},{"alias_kind":"pith_short_12","alias_value":"4EUURVWS5XN2","created_at":"2026-07-05T10:17:27.392562+00:00"},{"alias_kind":"pith_short_16","alias_value":"4EUURVWS5XN2ZIFZ","created_at":"2026-07-05T10:17:27.392562+00:00"},{"alias_kind":"pith_short_8","alias_value":"4EUURVWS","created_at":"2026-07-05T10:17:27.392562+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4EUURVWS5XN2ZIFZ3UTB66JPHK","json":"https://pith.science/pith/4EUURVWS5XN2ZIFZ3UTB66JPHK.json","graph_json":"https://pith.science/api/pith-number/4EUURVWS5XN2ZIFZ3UTB66JPHK/graph.json","events_json":"https://pith.science/api/pith-number/4EUURVWS5XN2ZIFZ3UTB66JPHK/events.json","paper":"https://pith.science/paper/4EUURVWS"},"agent_actions":{"view_html":"https://pith.science/pith/4EUURVWS5XN2ZIFZ3UTB66JPHK","download_json":"https://pith.science/pith/4EUURVWS5XN2ZIFZ3UTB66JPHK.json","view_paper":"https://pith.science/paper/4EUURVWS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.14486&json=true","fetch_graph":"https://pith.science/api/pith-number/4EUURVWS5XN2ZIFZ3UTB66JPHK/graph.json","fetch_events":"https://pith.science/api/pith-number/4EUURVWS5XN2ZIFZ3UTB66JPHK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4EUURVWS5XN2ZIFZ3UTB66JPHK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4EUURVWS5XN2ZIFZ3UTB66JPHK/action/storage_attestation","attest_author":"https://pith.science/pith/4EUURVWS5XN2ZIFZ3UTB66JPHK/action/author_attestation","sign_citation":"https://pith.science/pith/4EUURVWS5XN2ZIFZ3UTB66JPHK/action/citation_signature","submit_replication":"https://pith.science/pith/4EUURVWS5XN2ZIFZ3UTB66JPHK/action/replication_record"}},"created_at":"2026-07-05T10:17:27.392562+00:00","updated_at":"2026-07-05T10:17:27.392562+00:00"}