{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:V53WUQS3WFNNR7GYBOIHIV7WVY","short_pith_number":"pith:V53WUQS3","schema_version":"1.0","canonical_sha256":"af776a425bb15ad8fcd80b907457f6ae3a8bed1dba6b1c7fb5bb127763e5e53f","source":{"kind":"arxiv","id":"2409.15361","version":1},"attestation_state":"computed","paper":{"title":"Multitask Mayhem: Unveiling and Mitigating Safety Gaps in LLMs Fine-tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Essa Jan, Faizan Ahmad, Fareed Zaffar, Moiz Ali, Nouar AlDahoul, Yasir Zaki","submitted_at":"2024-09-18T08:04:24Z","abstract_excerpt":"Recent breakthroughs in Large Language Models (LLMs) have led to their adoption across a wide range of tasks, ranging from code generation to machine translation and sentiment analysis, etc. Red teaming/Safety alignment efforts show that fine-tuning models on benign (non-harmful) data could compromise safety. However, it remains unclear to what extent this phenomenon is influenced by different variables, including fine-tuning task, model calibrations, etc. This paper explores the task-wise safety degradation due to fine-tuning on downstream tasks such as summarization, code generation, transla"},"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":"2409.15361","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-09-18T08:04:24Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"aa2eb8814c3e3e8f024a7d2904813d238c3f925659b4329cf1616eeb93eeb244","abstract_canon_sha256":"af17e45b107a0084007b5a4201780eedec87f5892f0101f6bacd54d4078eddcb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:10:53.771605Z","signature_b64":"fI9hxHCLSguO/lXa2s6o18A0zQJ07dsjJ5LD5s0gZZYTDWPYprPCZY5wZazreNhpGygOmSMt51/+tcOiRcVRCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"af776a425bb15ad8fcd80b907457f6ae3a8bed1dba6b1c7fb5bb127763e5e53f","last_reissued_at":"2026-07-05T09:10:53.771177Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:10:53.771177Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multitask Mayhem: Unveiling and Mitigating Safety Gaps in LLMs Fine-tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Essa Jan, Faizan Ahmad, Fareed Zaffar, Moiz Ali, Nouar AlDahoul, Yasir Zaki","submitted_at":"2024-09-18T08:04:24Z","abstract_excerpt":"Recent breakthroughs in Large Language Models (LLMs) have led to their adoption across a wide range of tasks, ranging from code generation to machine translation and sentiment analysis, etc. Red teaming/Safety alignment efforts show that fine-tuning models on benign (non-harmful) data could compromise safety. However, it remains unclear to what extent this phenomenon is influenced by different variables, including fine-tuning task, model calibrations, etc. This paper explores the task-wise safety degradation due to fine-tuning on downstream tasks such as summarization, code generation, transla"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.15361","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/2409.15361/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":"2409.15361","created_at":"2026-07-05T09:10:53.771232+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.15361v1","created_at":"2026-07-05T09:10:53.771232+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.15361","created_at":"2026-07-05T09:10:53.771232+00:00"},{"alias_kind":"pith_short_12","alias_value":"V53WUQS3WFNN","created_at":"2026-07-05T09:10:53.771232+00:00"},{"alias_kind":"pith_short_16","alias_value":"V53WUQS3WFNNR7GY","created_at":"2026-07-05T09:10:53.771232+00:00"},{"alias_kind":"pith_short_8","alias_value":"V53WUQS3","created_at":"2026-07-05T09:10:53.771232+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.00686","citing_title":"Dialectics of Alignment: Harnessing Unsafe Knowledge for Dynamic Safety Routing","ref_index":37,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/V53WUQS3WFNNR7GYBOIHIV7WVY","json":"https://pith.science/pith/V53WUQS3WFNNR7GYBOIHIV7WVY.json","graph_json":"https://pith.science/api/pith-number/V53WUQS3WFNNR7GYBOIHIV7WVY/graph.json","events_json":"https://pith.science/api/pith-number/V53WUQS3WFNNR7GYBOIHIV7WVY/events.json","paper":"https://pith.science/paper/V53WUQS3"},"agent_actions":{"view_html":"https://pith.science/pith/V53WUQS3WFNNR7GYBOIHIV7WVY","download_json":"https://pith.science/pith/V53WUQS3WFNNR7GYBOIHIV7WVY.json","view_paper":"https://pith.science/paper/V53WUQS3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.15361&json=true","fetch_graph":"https://pith.science/api/pith-number/V53WUQS3WFNNR7GYBOIHIV7WVY/graph.json","fetch_events":"https://pith.science/api/pith-number/V53WUQS3WFNNR7GYBOIHIV7WVY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/V53WUQS3WFNNR7GYBOIHIV7WVY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/V53WUQS3WFNNR7GYBOIHIV7WVY/action/storage_attestation","attest_author":"https://pith.science/pith/V53WUQS3WFNNR7GYBOIHIV7WVY/action/author_attestation","sign_citation":"https://pith.science/pith/V53WUQS3WFNNR7GYBOIHIV7WVY/action/citation_signature","submit_replication":"https://pith.science/pith/V53WUQS3WFNNR7GYBOIHIV7WVY/action/replication_record"}},"created_at":"2026-07-05T09:10:53.771232+00:00","updated_at":"2026-07-05T09:10:53.771232+00:00"}