{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:P3SQX66CK2YA3FQP2NTY7LANMP","short_pith_number":"pith:P3SQX66C","schema_version":"1.0","canonical_sha256":"7ee50bfbc256b00d960fd3678fac0d63cf146145c93bd5dc70b42f46fb2622fd","source":{"kind":"arxiv","id":"2402.01109","version":6},"attestation_state":"computed","paper":{"title":"Vaccine: Perturbation-aware Alignment for Large Language Models against Harmful Fine-tuning Attack","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Ling Liu, Sihao Hu, Tiansheng Huang","submitted_at":"2024-02-02T02:56:50Z","abstract_excerpt":"The new paradigm of finetuning-as-a-service introduces a new attack surface for Large Language Models (LLMs): a few harmful data uploaded by users can easily trick the finetuning to produce an alignment-broken model. We conduct an empirical analysis and uncover a \\textit{harmful embedding drift} phenomenon, showing a probable cause of the alignment-broken effect. Inspired by our findings, we propose Vaccine, a perturbation-aware alignment technique to mitigate the security risk of users finetuning. The core idea of Vaccine is to produce invariant hidden embeddings by progressively adding craft"},"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":"2402.01109","kind":"arxiv","version":6},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-02-02T02:56:50Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"d69788c3c0f9bda340ffcf49b1ddd73b31f922ecf67220ae64a42c7229d4bcbb","abstract_canon_sha256":"c4253add02d69d88900e8ee5aa3577cb35c6d6dd60e93625a1d54aa1a6240f4b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:39:52.925899Z","signature_b64":"kJ9L9e6GGJwFffsC1hXHHXRioCJMwHX/7ACtB+6a4WhaYIdU3CZ6xw3xgPnMIMFw2hWBO+d5wFEzlBoRQbceBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7ee50bfbc256b00d960fd3678fac0d63cf146145c93bd5dc70b42f46fb2622fd","last_reissued_at":"2026-07-05T09:39:52.925409Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:39:52.925409Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Vaccine: Perturbation-aware Alignment for Large Language Models against Harmful Fine-tuning Attack","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Ling Liu, Sihao Hu, Tiansheng Huang","submitted_at":"2024-02-02T02:56:50Z","abstract_excerpt":"The new paradigm of finetuning-as-a-service introduces a new attack surface for Large Language Models (LLMs): a few harmful data uploaded by users can easily trick the finetuning to produce an alignment-broken model. We conduct an empirical analysis and uncover a \\textit{harmful embedding drift} phenomenon, showing a probable cause of the alignment-broken effect. Inspired by our findings, we propose Vaccine, a perturbation-aware alignment technique to mitigate the security risk of users finetuning. The core idea of Vaccine is to produce invariant hidden embeddings by progressively adding craft"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.01109","kind":"arxiv","version":6},"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/2402.01109/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":"2402.01109","created_at":"2026-07-05T09:39:52.925467+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.01109v6","created_at":"2026-07-05T09:39:52.925467+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.01109","created_at":"2026-07-05T09:39:52.925467+00:00"},{"alias_kind":"pith_short_12","alias_value":"P3SQX66CK2YA","created_at":"2026-07-05T09:39:52.925467+00:00"},{"alias_kind":"pith_short_16","alias_value":"P3SQX66CK2YA3FQP","created_at":"2026-07-05T09:39:52.925467+00:00"},{"alias_kind":"pith_short_8","alias_value":"P3SQX66C","created_at":"2026-07-05T09:39:52.925467+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2409.18169","citing_title":"Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey","ref_index":68,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/P3SQX66CK2YA3FQP2NTY7LANMP","json":"https://pith.science/pith/P3SQX66CK2YA3FQP2NTY7LANMP.json","graph_json":"https://pith.science/api/pith-number/P3SQX66CK2YA3FQP2NTY7LANMP/graph.json","events_json":"https://pith.science/api/pith-number/P3SQX66CK2YA3FQP2NTY7LANMP/events.json","paper":"https://pith.science/paper/P3SQX66C"},"agent_actions":{"view_html":"https://pith.science/pith/P3SQX66CK2YA3FQP2NTY7LANMP","download_json":"https://pith.science/pith/P3SQX66CK2YA3FQP2NTY7LANMP.json","view_paper":"https://pith.science/paper/P3SQX66C","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.01109&json=true","fetch_graph":"https://pith.science/api/pith-number/P3SQX66CK2YA3FQP2NTY7LANMP/graph.json","fetch_events":"https://pith.science/api/pith-number/P3SQX66CK2YA3FQP2NTY7LANMP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/P3SQX66CK2YA3FQP2NTY7LANMP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/P3SQX66CK2YA3FQP2NTY7LANMP/action/storage_attestation","attest_author":"https://pith.science/pith/P3SQX66CK2YA3FQP2NTY7LANMP/action/author_attestation","sign_citation":"https://pith.science/pith/P3SQX66CK2YA3FQP2NTY7LANMP/action/citation_signature","submit_replication":"https://pith.science/pith/P3SQX66CK2YA3FQP2NTY7LANMP/action/replication_record"}},"created_at":"2026-07-05T09:39:52.925467+00:00","updated_at":"2026-07-05T09:39:52.925467+00:00"}