{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SSE4HMBJH37A3UWRKAJBLYZ4WG","short_pith_number":"pith:SSE4HMBJ","schema_version":"1.0","canonical_sha256":"9489c3b0293efe0dd2d1501215e33cb18e5bccc90b1d80511fa1578f239e9984","source":{"kind":"arxiv","id":"2506.17209","version":1},"attestation_state":"computed","paper":{"title":"Fine-Tuning Lowers Safety and Disrupts Evaluation Consistency","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hillary Dawkins, Isar Nejadgholi, Kathleen C. Fraser, Svetlana Kiritchenko","submitted_at":"2025-06-20T17:57:12Z","abstract_excerpt":"Fine-tuning a general-purpose large language model (LLM) for a specific domain or task has become a routine procedure for ordinary users. However, fine-tuning is known to remove the safety alignment features of the model, even when the fine-tuning data does not contain any harmful content. We consider this to be a critical failure mode of LLMs due to the widespread uptake of fine-tuning, combined with the benign nature of the \"attack\". Most well-intentioned developers are likely unaware that they are deploying an LLM with reduced safety. On the other hand, this known vulnerability can be easil"},"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":"2506.17209","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-20T17:57:12Z","cross_cats_sorted":[],"title_canon_sha256":"0a7772da67e976082a0a5708ee5c45c9cc174e515edbf39bd943f901c0dc7318","abstract_canon_sha256":"4f4f6487897f9f10b27c829909e46ccd33e8bd5dfdabc66981cea2fe142cc892"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:24:58.230846Z","signature_b64":"eczfTRMm/ZRzvXC4BanqEp426JlR7T9DY3HyIhf03lzeYYE+DA60W0qK5WnoWskqsWzYl3ALnmi1nqQ/O8p5CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9489c3b0293efe0dd2d1501215e33cb18e5bccc90b1d80511fa1578f239e9984","last_reissued_at":"2026-07-05T11:24:58.230305Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:24:58.230305Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fine-Tuning Lowers Safety and Disrupts Evaluation Consistency","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hillary Dawkins, Isar Nejadgholi, Kathleen C. Fraser, Svetlana Kiritchenko","submitted_at":"2025-06-20T17:57:12Z","abstract_excerpt":"Fine-tuning a general-purpose large language model (LLM) for a specific domain or task has become a routine procedure for ordinary users. However, fine-tuning is known to remove the safety alignment features of the model, even when the fine-tuning data does not contain any harmful content. We consider this to be a critical failure mode of LLMs due to the widespread uptake of fine-tuning, combined with the benign nature of the \"attack\". Most well-intentioned developers are likely unaware that they are deploying an LLM with reduced safety. On the other hand, this known vulnerability can be easil"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.17209","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/2506.17209/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":"2506.17209","created_at":"2026-07-05T11:24:58.230382+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.17209v1","created_at":"2026-07-05T11:24:58.230382+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.17209","created_at":"2026-07-05T11:24:58.230382+00:00"},{"alias_kind":"pith_short_12","alias_value":"SSE4HMBJH37A","created_at":"2026-07-05T11:24:58.230382+00:00"},{"alias_kind":"pith_short_16","alias_value":"SSE4HMBJH37A3UWR","created_at":"2026-07-05T11:24:58.230382+00:00"},{"alias_kind":"pith_short_8","alias_value":"SSE4HMBJ","created_at":"2026-07-05T11:24:58.230382+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.14418","citing_title":"The Great Pretender: A Stochasticity Problem in LLM Jailbreak","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SSE4HMBJH37A3UWRKAJBLYZ4WG","json":"https://pith.science/pith/SSE4HMBJH37A3UWRKAJBLYZ4WG.json","graph_json":"https://pith.science/api/pith-number/SSE4HMBJH37A3UWRKAJBLYZ4WG/graph.json","events_json":"https://pith.science/api/pith-number/SSE4HMBJH37A3UWRKAJBLYZ4WG/events.json","paper":"https://pith.science/paper/SSE4HMBJ"},"agent_actions":{"view_html":"https://pith.science/pith/SSE4HMBJH37A3UWRKAJBLYZ4WG","download_json":"https://pith.science/pith/SSE4HMBJH37A3UWRKAJBLYZ4WG.json","view_paper":"https://pith.science/paper/SSE4HMBJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.17209&json=true","fetch_graph":"https://pith.science/api/pith-number/SSE4HMBJH37A3UWRKAJBLYZ4WG/graph.json","fetch_events":"https://pith.science/api/pith-number/SSE4HMBJH37A3UWRKAJBLYZ4WG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SSE4HMBJH37A3UWRKAJBLYZ4WG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SSE4HMBJH37A3UWRKAJBLYZ4WG/action/storage_attestation","attest_author":"https://pith.science/pith/SSE4HMBJH37A3UWRKAJBLYZ4WG/action/author_attestation","sign_citation":"https://pith.science/pith/SSE4HMBJH37A3UWRKAJBLYZ4WG/action/citation_signature","submit_replication":"https://pith.science/pith/SSE4HMBJH37A3UWRKAJBLYZ4WG/action/replication_record"}},"created_at":"2026-07-05T11:24:58.230382+00:00","updated_at":"2026-07-05T11:24:58.230382+00:00"}