{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:4MTKL7CL53P4OA7STXULTIO6KV","short_pith_number":"pith:4MTKL7CL","schema_version":"1.0","canonical_sha256":"e326a5fc4beedfc703f29de8b9a1de555a2c83ec45a2fa8c8d603a778db12dae","source":{"kind":"arxiv","id":"2405.13820","version":2},"attestation_state":"computed","paper":{"title":"Towards Comprehensive Post Safety Alignment of Large Language Models via Safety Patching","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bing Qin, Jiahe Guo, Tat-Seng Chua, Ting Liu, Weixiang Zhao, Xingyu Sui, Yang Deng, Yanyan Zhao, Yulin Hu, Zhuojun Li","submitted_at":"2024-05-22T16:51:07Z","abstract_excerpt":"Safety alignment of large language models (LLMs) has been gaining increasing attention. However, current safety-aligned LLMs suffer from the fragile and imbalanced safety mechanisms, which can still be induced to generate unsafe responses, exhibit over-safety by rejecting safe user inputs, and fail to preserve general utility after safety alignment. To this end, we propose a novel post safety alignment (PSA) method to address these inherent and emerging safety challenges, including safety enhancement, over-safety mitigation, and utility preservation. In specific, we introduce \\textsc{SafePatch"},"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":"2405.13820","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-05-22T16:51:07Z","cross_cats_sorted":[],"title_canon_sha256":"b5ea66ea0da5e85128fe9172ea69ec7635fddf8980addcfd1e7767deef128b00","abstract_canon_sha256":"d208a78f745035370726c25082dc6401957e07d7c203b998536243dc5db75642"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:50:02.588092Z","signature_b64":"625dBwaeLXYTmIk1fLHOAGh5HjwLT6uN3i9KbeyuUMOeNsIzTX2iDiUSdI486zQ4U+81CrCeumB9ZnJ2ZRhWBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e326a5fc4beedfc703f29de8b9a1de555a2c83ec45a2fa8c8d603a778db12dae","last_reissued_at":"2026-07-05T09:50:02.587607Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:50:02.587607Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Comprehensive Post Safety Alignment of Large Language Models via Safety Patching","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bing Qin, Jiahe Guo, Tat-Seng Chua, Ting Liu, Weixiang Zhao, Xingyu Sui, Yang Deng, Yanyan Zhao, Yulin Hu, Zhuojun Li","submitted_at":"2024-05-22T16:51:07Z","abstract_excerpt":"Safety alignment of large language models (LLMs) has been gaining increasing attention. However, current safety-aligned LLMs suffer from the fragile and imbalanced safety mechanisms, which can still be induced to generate unsafe responses, exhibit over-safety by rejecting safe user inputs, and fail to preserve general utility after safety alignment. To this end, we propose a novel post safety alignment (PSA) method to address these inherent and emerging safety challenges, including safety enhancement, over-safety mitigation, and utility preservation. In specific, we introduce \\textsc{SafePatch"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.13820","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/2405.13820/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":"2405.13820","created_at":"2026-07-05T09:50:02.587669+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.13820v2","created_at":"2026-07-05T09:50:02.587669+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.13820","created_at":"2026-07-05T09:50:02.587669+00:00"},{"alias_kind":"pith_short_12","alias_value":"4MTKL7CL53P4","created_at":"2026-07-05T09:50:02.587669+00:00"},{"alias_kind":"pith_short_16","alias_value":"4MTKL7CL53P4OA7S","created_at":"2026-07-05T09:50:02.587669+00:00"},{"alias_kind":"pith_short_8","alias_value":"4MTKL7CL","created_at":"2026-07-05T09:50:02.587669+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.18104","citing_title":"Safety Geometry Collapse in Multimodal LLMs and Adaptive Drift Correction","ref_index":48,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4MTKL7CL53P4OA7STXULTIO6KV","json":"https://pith.science/pith/4MTKL7CL53P4OA7STXULTIO6KV.json","graph_json":"https://pith.science/api/pith-number/4MTKL7CL53P4OA7STXULTIO6KV/graph.json","events_json":"https://pith.science/api/pith-number/4MTKL7CL53P4OA7STXULTIO6KV/events.json","paper":"https://pith.science/paper/4MTKL7CL"},"agent_actions":{"view_html":"https://pith.science/pith/4MTKL7CL53P4OA7STXULTIO6KV","download_json":"https://pith.science/pith/4MTKL7CL53P4OA7STXULTIO6KV.json","view_paper":"https://pith.science/paper/4MTKL7CL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.13820&json=true","fetch_graph":"https://pith.science/api/pith-number/4MTKL7CL53P4OA7STXULTIO6KV/graph.json","fetch_events":"https://pith.science/api/pith-number/4MTKL7CL53P4OA7STXULTIO6KV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4MTKL7CL53P4OA7STXULTIO6KV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4MTKL7CL53P4OA7STXULTIO6KV/action/storage_attestation","attest_author":"https://pith.science/pith/4MTKL7CL53P4OA7STXULTIO6KV/action/author_attestation","sign_citation":"https://pith.science/pith/4MTKL7CL53P4OA7STXULTIO6KV/action/citation_signature","submit_replication":"https://pith.science/pith/4MTKL7CL53P4OA7STXULTIO6KV/action/replication_record"}},"created_at":"2026-07-05T09:50:02.587669+00:00","updated_at":"2026-07-05T09:50:02.587669+00:00"}