{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:XESEPPKD7BTKUFWCYFSA3Z6RF7","short_pith_number":"pith:XESEPPKD","schema_version":"1.0","canonical_sha256":"b92447bd43f866aa16c2c1640de7d12ff8b37b08a1e1d3bb4dd3f9fa0a5864d0","source":{"kind":"arxiv","id":"2409.00598","version":2},"attestation_state":"computed","paper":{"title":"Automatic Pseudo-Harmful Prompt Generation for Evaluating False Refusals in Large Language Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CR","cs.CY","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bang An, Furong Huang, Michael-Andrei Panaitescu-Liess, Ruiyi Zhang, Sicheng Zhu, Yuancheng Xu","submitted_at":"2024-09-01T03:25:59Z","abstract_excerpt":"Safety-aligned large language models (LLMs) sometimes falsely refuse pseudo-harmful prompts, like \"how to kill a mosquito,\" which are actually harmless. Frequent false refusals not only frustrate users but also provoke a public backlash against the very values alignment seeks to protect. In this paper, we propose the first method to auto-generate diverse, content-controlled, and model-dependent pseudo-harmful prompts. Using this method, we construct an evaluation dataset called PHTest, which is ten times larger than existing datasets, covers more false refusal patterns, and separately labels c"},"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":"2409.00598","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-09-01T03:25:59Z","cross_cats_sorted":["cs.CR","cs.CY","cs.LG"],"title_canon_sha256":"645b749c5e3ec5b0d41887ea74349e17cd5e5785d63a5debeca74ce0b96a93b2","abstract_canon_sha256":"faa12810232d930fd62f80e1adf56780c8258f30efaa431ff912c05e99eb4034"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:20.914739Z","signature_b64":"cBvWteQ/I1SZpD83no0EHnATrE96t2uon0eiWrcOKoN0X9Ng9AUor75f0bWvFcT83XayFXJ/EMgBdL/CexPbAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b92447bd43f866aa16c2c1640de7d12ff8b37b08a1e1d3bb4dd3f9fa0a5864d0","last_reissued_at":"2026-07-05T11:19:20.914282Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:20.914282Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Automatic Pseudo-Harmful Prompt Generation for Evaluating False Refusals in Large Language Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CR","cs.CY","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bang An, Furong Huang, Michael-Andrei Panaitescu-Liess, Ruiyi Zhang, Sicheng Zhu, Yuancheng Xu","submitted_at":"2024-09-01T03:25:59Z","abstract_excerpt":"Safety-aligned large language models (LLMs) sometimes falsely refuse pseudo-harmful prompts, like \"how to kill a mosquito,\" which are actually harmless. Frequent false refusals not only frustrate users but also provoke a public backlash against the very values alignment seeks to protect. In this paper, we propose the first method to auto-generate diverse, content-controlled, and model-dependent pseudo-harmful prompts. Using this method, we construct an evaluation dataset called PHTest, which is ten times larger than existing datasets, covers more false refusal patterns, and separately labels c"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.00598","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/2409.00598/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":"2409.00598","created_at":"2026-07-05T11:19:20.914335+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.00598v2","created_at":"2026-07-05T11:19:20.914335+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.00598","created_at":"2026-07-05T11:19:20.914335+00:00"},{"alias_kind":"pith_short_12","alias_value":"XESEPPKD7BTK","created_at":"2026-07-05T11:19:20.914335+00:00"},{"alias_kind":"pith_short_16","alias_value":"XESEPPKD7BTKUFWC","created_at":"2026-07-05T11:19:20.914335+00:00"},{"alias_kind":"pith_short_8","alias_value":"XESEPPKD","created_at":"2026-07-05T11:19:20.914335+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06196","citing_title":"Pluralis v0.1: Towards a Multicultural, Multimodal, Multilingual Benchmark for AI Risk and Reliability","ref_index":65,"is_internal_anchor":true},{"citing_arxiv_id":"2606.21147","citing_title":"AOR-Bench: Do Large Audio Language Models Over-Refuse Pseudo-Harmful Queries?","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XESEPPKD7BTKUFWCYFSA3Z6RF7","json":"https://pith.science/pith/XESEPPKD7BTKUFWCYFSA3Z6RF7.json","graph_json":"https://pith.science/api/pith-number/XESEPPKD7BTKUFWCYFSA3Z6RF7/graph.json","events_json":"https://pith.science/api/pith-number/XESEPPKD7BTKUFWCYFSA3Z6RF7/events.json","paper":"https://pith.science/paper/XESEPPKD"},"agent_actions":{"view_html":"https://pith.science/pith/XESEPPKD7BTKUFWCYFSA3Z6RF7","download_json":"https://pith.science/pith/XESEPPKD7BTKUFWCYFSA3Z6RF7.json","view_paper":"https://pith.science/paper/XESEPPKD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.00598&json=true","fetch_graph":"https://pith.science/api/pith-number/XESEPPKD7BTKUFWCYFSA3Z6RF7/graph.json","fetch_events":"https://pith.science/api/pith-number/XESEPPKD7BTKUFWCYFSA3Z6RF7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XESEPPKD7BTKUFWCYFSA3Z6RF7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XESEPPKD7BTKUFWCYFSA3Z6RF7/action/storage_attestation","attest_author":"https://pith.science/pith/XESEPPKD7BTKUFWCYFSA3Z6RF7/action/author_attestation","sign_citation":"https://pith.science/pith/XESEPPKD7BTKUFWCYFSA3Z6RF7/action/citation_signature","submit_replication":"https://pith.science/pith/XESEPPKD7BTKUFWCYFSA3Z6RF7/action/replication_record"}},"created_at":"2026-07-05T11:19:20.914335+00:00","updated_at":"2026-07-05T11:19:20.914335+00:00"}