{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:F22JBBSLPJGDSTLZFNQI2WCWUO","short_pith_number":"pith:F22JBBSL","schema_version":"1.0","canonical_sha256":"2eb490864b7a4c394d792b608d5856a39f2aa26739eea05a32d2d2fe11c5e75f","source":{"kind":"arxiv","id":"2405.15202","version":1},"attestation_state":"computed","paper":{"title":"Cross-Task Defense: Instruction-Tuning LLMs for Content Safety","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.CL","authors_text":"Aichi Chien, Evangelos Papalexakis, Jia Chen, Jiachen Li, Wen Xiao, Yue Dong, Yu Fu","submitted_at":"2024-05-24T04:14:32Z","abstract_excerpt":"Recent studies reveal that Large Language Models (LLMs) face challenges in balancing safety with utility, particularly when processing long texts for NLP tasks like summarization and translation. Despite defenses against malicious short questions, the ability of LLMs to safely handle dangerous long content, such as manuals teaching illicit activities, remains unclear. Our work aims to develop robust defenses for LLMs in processing malicious documents alongside benign NLP task queries. We introduce a defense dataset comprised of safety-related examples and propose single-task and mixed-task los"},"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":"2405.15202","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-05-24T04:14:32Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"3a204891bf039ee65d8e5c9beeb1eb92dd0ab3249c2b2c6cbd8dcd158010a63c","abstract_canon_sha256":"b47b1f37d446169e498d172b4470024d6faae1996a154dddf12d8a04b612d799"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:22:46.367110Z","signature_b64":"0xDvw90ge+Q0eXo71pJhJ/afrYQQbH//jQLSPPfJMPjKNn5sWR02QBpb0h1BPbJKOgFvL1cO+7/+sKqud4uHBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2eb490864b7a4c394d792b608d5856a39f2aa26739eea05a32d2d2fe11c5e75f","last_reissued_at":"2026-07-05T08:22:46.366625Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:22:46.366625Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Cross-Task Defense: Instruction-Tuning LLMs for Content Safety","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.CL","authors_text":"Aichi Chien, Evangelos Papalexakis, Jia Chen, Jiachen Li, Wen Xiao, Yue Dong, Yu Fu","submitted_at":"2024-05-24T04:14:32Z","abstract_excerpt":"Recent studies reveal that Large Language Models (LLMs) face challenges in balancing safety with utility, particularly when processing long texts for NLP tasks like summarization and translation. Despite defenses against malicious short questions, the ability of LLMs to safely handle dangerous long content, such as manuals teaching illicit activities, remains unclear. Our work aims to develop robust defenses for LLMs in processing malicious documents alongside benign NLP task queries. We introduce a defense dataset comprised of safety-related examples and propose single-task and mixed-task los"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.15202","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/2405.15202/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":"2405.15202","created_at":"2026-07-05T08:22:46.366684+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.15202v1","created_at":"2026-07-05T08:22:46.366684+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.15202","created_at":"2026-07-05T08:22:46.366684+00:00"},{"alias_kind":"pith_short_12","alias_value":"F22JBBSLPJGD","created_at":"2026-07-05T08:22:46.366684+00:00"},{"alias_kind":"pith_short_16","alias_value":"F22JBBSLPJGDSTLZ","created_at":"2026-07-05T08:22:46.366684+00:00"},{"alias_kind":"pith_short_8","alias_value":"F22JBBSL","created_at":"2026-07-05T08:22:46.366684+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2412.17686","citing_title":"Large Language Model Safety: A Holistic Survey","ref_index":13,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/F22JBBSLPJGDSTLZFNQI2WCWUO","json":"https://pith.science/pith/F22JBBSLPJGDSTLZFNQI2WCWUO.json","graph_json":"https://pith.science/api/pith-number/F22JBBSLPJGDSTLZFNQI2WCWUO/graph.json","events_json":"https://pith.science/api/pith-number/F22JBBSLPJGDSTLZFNQI2WCWUO/events.json","paper":"https://pith.science/paper/F22JBBSL"},"agent_actions":{"view_html":"https://pith.science/pith/F22JBBSLPJGDSTLZFNQI2WCWUO","download_json":"https://pith.science/pith/F22JBBSLPJGDSTLZFNQI2WCWUO.json","view_paper":"https://pith.science/paper/F22JBBSL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.15202&json=true","fetch_graph":"https://pith.science/api/pith-number/F22JBBSLPJGDSTLZFNQI2WCWUO/graph.json","fetch_events":"https://pith.science/api/pith-number/F22JBBSLPJGDSTLZFNQI2WCWUO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/F22JBBSLPJGDSTLZFNQI2WCWUO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/F22JBBSLPJGDSTLZFNQI2WCWUO/action/storage_attestation","attest_author":"https://pith.science/pith/F22JBBSLPJGDSTLZFNQI2WCWUO/action/author_attestation","sign_citation":"https://pith.science/pith/F22JBBSLPJGDSTLZFNQI2WCWUO/action/citation_signature","submit_replication":"https://pith.science/pith/F22JBBSLPJGDSTLZFNQI2WCWUO/action/replication_record"}},"created_at":"2026-07-05T08:22:46.366684+00:00","updated_at":"2026-07-05T08:22:46.366684+00:00"}