{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:JYQ2R2CGOWXDXZCF2R7YPIMSL5","short_pith_number":"pith:JYQ2R2CG","schema_version":"1.0","canonical_sha256":"4e21a8e84675ae3be445d47f87a1925f50fc0fc1f317f36b33e93efd9d506a79","source":{"kind":"arxiv","id":"2506.17353","version":1},"attestation_state":"computed","paper":{"title":"Differentiation-Based Extraction of Proprietary Data from Fine-Tuned LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Daoyuan Wu, Shuai Wang, Zhendong Su, Zongjie Li","submitted_at":"2025-06-20T02:43:36Z","abstract_excerpt":"The increasing demand for domain-specific and human-aligned Large Language Models (LLMs) has led to the widespread adoption of Supervised Fine-Tuning (SFT) techniques. SFT datasets often comprise valuable instruction-response pairs, making them highly valuable targets for potential extraction. This paper studies this critical research problem for the first time. We start by formally defining and formulating the problem, then explore various attack goals, types, and variants based on the unique properties of SFT data in real-world scenarios. Based on our analysis of extraction behaviors of dire"},"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":"2506.17353","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-06-20T02:43:36Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"d9d3fae6a668bfd84d7ccdaf346ca1f80230494b7ce664dad1a3e0264f9802d1","abstract_canon_sha256":"9ead9ca0ba6a7102722a62595479026d84f0a5f61c07bff7f1344938786da282"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:25:16.446034Z","signature_b64":"X7jiHttLVHBZKARGCaT67yM1ci9dCP4f+p0kpjOQvW9VuOuOF8ykxsMoVsz6h+DLtR/q4NeuKo3ujO3R9LNKCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4e21a8e84675ae3be445d47f87a1925f50fc0fc1f317f36b33e93efd9d506a79","last_reissued_at":"2026-07-05T11:25:16.445571Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:25:16.445571Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Differentiation-Based Extraction of Proprietary Data from Fine-Tuned LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Daoyuan Wu, Shuai Wang, Zhendong Su, Zongjie Li","submitted_at":"2025-06-20T02:43:36Z","abstract_excerpt":"The increasing demand for domain-specific and human-aligned Large Language Models (LLMs) has led to the widespread adoption of Supervised Fine-Tuning (SFT) techniques. SFT datasets often comprise valuable instruction-response pairs, making them highly valuable targets for potential extraction. This paper studies this critical research problem for the first time. We start by formally defining and formulating the problem, then explore various attack goals, types, and variants based on the unique properties of SFT data in real-world scenarios. Based on our analysis of extraction behaviors of dire"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.17353","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/2506.17353/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":"2506.17353","created_at":"2026-07-05T11:25:16.445630+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.17353v1","created_at":"2026-07-05T11:25:16.445630+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.17353","created_at":"2026-07-05T11:25:16.445630+00:00"},{"alias_kind":"pith_short_12","alias_value":"JYQ2R2CGOWXD","created_at":"2026-07-05T11:25:16.445630+00:00"},{"alias_kind":"pith_short_16","alias_value":"JYQ2R2CGOWXDXZCF","created_at":"2026-07-05T11:25:16.445630+00:00"},{"alias_kind":"pith_short_8","alias_value":"JYQ2R2CG","created_at":"2026-07-05T11:25:16.445630+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/JYQ2R2CGOWXDXZCF2R7YPIMSL5","json":"https://pith.science/pith/JYQ2R2CGOWXDXZCF2R7YPIMSL5.json","graph_json":"https://pith.science/api/pith-number/JYQ2R2CGOWXDXZCF2R7YPIMSL5/graph.json","events_json":"https://pith.science/api/pith-number/JYQ2R2CGOWXDXZCF2R7YPIMSL5/events.json","paper":"https://pith.science/paper/JYQ2R2CG"},"agent_actions":{"view_html":"https://pith.science/pith/JYQ2R2CGOWXDXZCF2R7YPIMSL5","download_json":"https://pith.science/pith/JYQ2R2CGOWXDXZCF2R7YPIMSL5.json","view_paper":"https://pith.science/paper/JYQ2R2CG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.17353&json=true","fetch_graph":"https://pith.science/api/pith-number/JYQ2R2CGOWXDXZCF2R7YPIMSL5/graph.json","fetch_events":"https://pith.science/api/pith-number/JYQ2R2CGOWXDXZCF2R7YPIMSL5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JYQ2R2CGOWXDXZCF2R7YPIMSL5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JYQ2R2CGOWXDXZCF2R7YPIMSL5/action/storage_attestation","attest_author":"https://pith.science/pith/JYQ2R2CGOWXDXZCF2R7YPIMSL5/action/author_attestation","sign_citation":"https://pith.science/pith/JYQ2R2CGOWXDXZCF2R7YPIMSL5/action/citation_signature","submit_replication":"https://pith.science/pith/JYQ2R2CGOWXDXZCF2R7YPIMSL5/action/replication_record"}},"created_at":"2026-07-05T11:25:16.445630+00:00","updated_at":"2026-07-05T11:25:16.445630+00:00"}