{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:R54R5QGSWRWA5A3BCKX3NLD23D","short_pith_number":"pith:R54R5QGS","schema_version":"1.0","canonical_sha256":"8f791ec0d2b46c0e836112afb6ac7ad8d14b81d52041fd4662e3b0917fc8a9a9","source":{"kind":"arxiv","id":"2504.21036","version":2},"attestation_state":"computed","paper":{"title":"Can Differentially Private Fine-tuning LLMs Protect Against Privacy Attacks?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CR","authors_text":"Hao Du, Shang Liu, Yang Cao","submitted_at":"2025-04-28T05:34:53Z","abstract_excerpt":"Fine-tuning large language models (LLMs) has become an essential strategy for adapting them to specialized tasks; however, this process introduces significant privacy challenges, as sensitive training data may be inadvertently memorized and exposed. Although differential privacy (DP) offers strong theoretical guarantees against such leakage, its empirical privacy effectiveness on LLMs remains unclear, especially under different fine-tuning methods. In this paper, we systematically investigate the impact of DP across fine-tuning methods and privacy budgets, using both data extraction and member"},"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":"2504.21036","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-04-28T05:34:53Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"d5c3e8aff4a94c2632ad5921c6d9382f5ae613fd462dc73e31f23f43cf8e4696","abstract_canon_sha256":"f1584f0b028b58e5c4b19812be425e8b4d7239dd68ec562b764d1ffba6d28c0e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:56:39.924964Z","signature_b64":"2BscB5YuUzJrJNj7pon0sgDQhgFDsywfpudNcDXsXOneKizcRmJTVr4I658v9VMcEORCaOzPTbUdJkSbr6u+Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8f791ec0d2b46c0e836112afb6ac7ad8d14b81d52041fd4662e3b0917fc8a9a9","last_reissued_at":"2026-07-05T10:56:39.924514Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:56:39.924514Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Can Differentially Private Fine-tuning LLMs Protect Against Privacy Attacks?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CR","authors_text":"Hao Du, Shang Liu, Yang Cao","submitted_at":"2025-04-28T05:34:53Z","abstract_excerpt":"Fine-tuning large language models (LLMs) has become an essential strategy for adapting them to specialized tasks; however, this process introduces significant privacy challenges, as sensitive training data may be inadvertently memorized and exposed. Although differential privacy (DP) offers strong theoretical guarantees against such leakage, its empirical privacy effectiveness on LLMs remains unclear, especially under different fine-tuning methods. In this paper, we systematically investigate the impact of DP across fine-tuning methods and privacy budgets, using both data extraction and member"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.21036","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/2504.21036/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":"2504.21036","created_at":"2026-07-05T10:56:39.924566+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.21036v2","created_at":"2026-07-05T10:56:39.924566+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.21036","created_at":"2026-07-05T10:56:39.924566+00:00"},{"alias_kind":"pith_short_12","alias_value":"R54R5QGSWRWA","created_at":"2026-07-05T10:56:39.924566+00:00"},{"alias_kind":"pith_short_16","alias_value":"R54R5QGSWRWA5A3B","created_at":"2026-07-05T10:56:39.924566+00:00"},{"alias_kind":"pith_short_8","alias_value":"R54R5QGS","created_at":"2026-07-05T10:56:39.924566+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/R54R5QGSWRWA5A3BCKX3NLD23D","json":"https://pith.science/pith/R54R5QGSWRWA5A3BCKX3NLD23D.json","graph_json":"https://pith.science/api/pith-number/R54R5QGSWRWA5A3BCKX3NLD23D/graph.json","events_json":"https://pith.science/api/pith-number/R54R5QGSWRWA5A3BCKX3NLD23D/events.json","paper":"https://pith.science/paper/R54R5QGS"},"agent_actions":{"view_html":"https://pith.science/pith/R54R5QGSWRWA5A3BCKX3NLD23D","download_json":"https://pith.science/pith/R54R5QGSWRWA5A3BCKX3NLD23D.json","view_paper":"https://pith.science/paper/R54R5QGS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.21036&json=true","fetch_graph":"https://pith.science/api/pith-number/R54R5QGSWRWA5A3BCKX3NLD23D/graph.json","fetch_events":"https://pith.science/api/pith-number/R54R5QGSWRWA5A3BCKX3NLD23D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/R54R5QGSWRWA5A3BCKX3NLD23D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/R54R5QGSWRWA5A3BCKX3NLD23D/action/storage_attestation","attest_author":"https://pith.science/pith/R54R5QGSWRWA5A3BCKX3NLD23D/action/author_attestation","sign_citation":"https://pith.science/pith/R54R5QGSWRWA5A3BCKX3NLD23D/action/citation_signature","submit_replication":"https://pith.science/pith/R54R5QGSWRWA5A3BCKX3NLD23D/action/replication_record"}},"created_at":"2026-07-05T10:56:39.924566+00:00","updated_at":"2026-07-05T10:56:39.924566+00:00"}