{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:AY2RYV7V75NGA5QUTPZFGQLQQ4","short_pith_number":"pith:AY2RYV7V","schema_version":"1.0","canonical_sha256":"06351c57f5ff5a6076149bf2534170872e9c819859ec37f35302c6976886e7cc","source":{"kind":"arxiv","id":"2507.18631","version":2},"attestation_state":"computed","paper":{"title":"Layer-Aware Representation Filtering: Purifying Finetuning Data to Preserve LLM Safety Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Hao Li, Jing Shao, Lei Sha, Lijun Li, Rui Li, Xianyi Wei, Zhenghao Lu","submitted_at":"2025-07-24T17:59:24Z","abstract_excerpt":"With rapid advancement and increasing accessibility of LLMs, fine-tuning aligned models has become a critical step for adapting them to real-world applications, which makes the safety of this fine-tuning process more important than ever. However, recent studies have highlighted a critical challenge: even when fine-tuning with seemingly benign downstream datasets, the safety of aligned LLMs can be compromised, making them more susceptible to malicious instructions.\n  In this paper, we show that fine-tuning datasets often contain samples with safety-degrading features that are not easily identif"},"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":"2507.18631","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-07-24T17:59:24Z","cross_cats_sorted":[],"title_canon_sha256":"d8083ce27bc1bc58c0eb97399a408512142d74b1ba1837b1fccb201e4e435d77","abstract_canon_sha256":"1620b8ceaf4e5ccec2e117b7d5c93f0c39c3067ee927d9d4395ee2dc6511c1b5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:43:15.042120Z","signature_b64":"7XA+nPE+asv0FFlGn0AvKcPaYgsCP4ZvXR/jr3/aPu2qeiWUkinbDc9Z+xylEBtfu08qxIQ1v5eNLaJ65zyhAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"06351c57f5ff5a6076149bf2534170872e9c819859ec37f35302c6976886e7cc","last_reissued_at":"2026-07-05T11:43:15.041655Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:43:15.041655Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Layer-Aware Representation Filtering: Purifying Finetuning Data to Preserve LLM Safety Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Hao Li, Jing Shao, Lei Sha, Lijun Li, Rui Li, Xianyi Wei, Zhenghao Lu","submitted_at":"2025-07-24T17:59:24Z","abstract_excerpt":"With rapid advancement and increasing accessibility of LLMs, fine-tuning aligned models has become a critical step for adapting them to real-world applications, which makes the safety of this fine-tuning process more important than ever. However, recent studies have highlighted a critical challenge: even when fine-tuning with seemingly benign downstream datasets, the safety of aligned LLMs can be compromised, making them more susceptible to malicious instructions.\n  In this paper, we show that fine-tuning datasets often contain samples with safety-degrading features that are not easily identif"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.18631","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/2507.18631/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":"2507.18631","created_at":"2026-07-05T11:43:15.041724+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.18631v2","created_at":"2026-07-05T11:43:15.041724+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.18631","created_at":"2026-07-05T11:43:15.041724+00:00"},{"alias_kind":"pith_short_12","alias_value":"AY2RYV7V75NG","created_at":"2026-07-05T11:43:15.041724+00:00"},{"alias_kind":"pith_short_16","alias_value":"AY2RYV7V75NGA5QU","created_at":"2026-07-05T11:43:15.041724+00:00"},{"alias_kind":"pith_short_8","alias_value":"AY2RYV7V","created_at":"2026-07-05T11:43:15.041724+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/AY2RYV7V75NGA5QUTPZFGQLQQ4","json":"https://pith.science/pith/AY2RYV7V75NGA5QUTPZFGQLQQ4.json","graph_json":"https://pith.science/api/pith-number/AY2RYV7V75NGA5QUTPZFGQLQQ4/graph.json","events_json":"https://pith.science/api/pith-number/AY2RYV7V75NGA5QUTPZFGQLQQ4/events.json","paper":"https://pith.science/paper/AY2RYV7V"},"agent_actions":{"view_html":"https://pith.science/pith/AY2RYV7V75NGA5QUTPZFGQLQQ4","download_json":"https://pith.science/pith/AY2RYV7V75NGA5QUTPZFGQLQQ4.json","view_paper":"https://pith.science/paper/AY2RYV7V","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.18631&json=true","fetch_graph":"https://pith.science/api/pith-number/AY2RYV7V75NGA5QUTPZFGQLQQ4/graph.json","fetch_events":"https://pith.science/api/pith-number/AY2RYV7V75NGA5QUTPZFGQLQQ4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AY2RYV7V75NGA5QUTPZFGQLQQ4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AY2RYV7V75NGA5QUTPZFGQLQQ4/action/storage_attestation","attest_author":"https://pith.science/pith/AY2RYV7V75NGA5QUTPZFGQLQQ4/action/author_attestation","sign_citation":"https://pith.science/pith/AY2RYV7V75NGA5QUTPZFGQLQQ4/action/citation_signature","submit_replication":"https://pith.science/pith/AY2RYV7V75NGA5QUTPZFGQLQQ4/action/replication_record"}},"created_at":"2026-07-05T11:43:15.041724+00:00","updated_at":"2026-07-05T11:43:15.041724+00:00"}