{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:EGLX34FCUMWYMWTDRWTS3ITFVZ","short_pith_number":"pith:EGLX34FC","schema_version":"1.0","canonical_sha256":"21977df0a2a32d865a638da72da265ae72d3f23e2897f4771abb242142571b14","source":{"kind":"arxiv","id":"2405.18540","version":2},"attestation_state":"computed","paper":{"title":"Learning diverse attacks on large language models for robust red-teaming and safety tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"David Dobre, Gauthier Gidel, Juho Lee, Kenji Kawaguchi, Lynn Cherif, Minsu Kim, Moksh Jain, Nikolay Malkin, Seanie Lee, Sung Ju Hwang, Yoshua Bengio","submitted_at":"2024-05-28T19:16:17Z","abstract_excerpt":"Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing ap"},"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":"2405.18540","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-05-28T19:16:17Z","cross_cats_sorted":["cs.CR","cs.LG"],"title_canon_sha256":"593e5a7335985fae38678c90fc3c09982246292c010560667ec1d7b6f7c798cc","abstract_canon_sha256":"e2e70b0eecd10ebb7780a99c4453d414577ed65f4649cdb2665123ae57aae437"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:21:19.568620Z","signature_b64":"DJHXsyMYqY8QcQsrAibwa4eR0s1Y/9H82v4OMEGFpRp7qhO/kGRNg/KbbaxaMzwew/7ldvmxkOQROQj5DREMAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"21977df0a2a32d865a638da72da265ae72d3f23e2897f4771abb242142571b14","last_reissued_at":"2026-07-05T10:21:19.568097Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:21:19.568097Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning diverse attacks on large language models for robust red-teaming and safety tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"David Dobre, Gauthier Gidel, Juho Lee, Kenji Kawaguchi, Lynn Cherif, Minsu Kim, Moksh Jain, Nikolay Malkin, Seanie Lee, Sung Ju Hwang, Yoshua Bengio","submitted_at":"2024-05-28T19:16:17Z","abstract_excerpt":"Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing ap"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.18540","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/2405.18540/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":"2405.18540","created_at":"2026-07-05T10:21:19.568151+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.18540v2","created_at":"2026-07-05T10:21:19.568151+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.18540","created_at":"2026-07-05T10:21:19.568151+00:00"},{"alias_kind":"pith_short_12","alias_value":"EGLX34FCUMWY","created_at":"2026-07-05T10:21:19.568151+00:00"},{"alias_kind":"pith_short_16","alias_value":"EGLX34FCUMWYMWTD","created_at":"2026-07-05T10:21:19.568151+00:00"},{"alias_kind":"pith_short_8","alias_value":"EGLX34FC","created_at":"2026-07-05T10:21:19.568151+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.00553","citing_title":"Stable-GFlowNet: Toward Diverse and Robust LLM Red-Teaming via Contrastive Trajectory Balance","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EGLX34FCUMWYMWTDRWTS3ITFVZ","json":"https://pith.science/pith/EGLX34FCUMWYMWTDRWTS3ITFVZ.json","graph_json":"https://pith.science/api/pith-number/EGLX34FCUMWYMWTDRWTS3ITFVZ/graph.json","events_json":"https://pith.science/api/pith-number/EGLX34FCUMWYMWTDRWTS3ITFVZ/events.json","paper":"https://pith.science/paper/EGLX34FC"},"agent_actions":{"view_html":"https://pith.science/pith/EGLX34FCUMWYMWTDRWTS3ITFVZ","download_json":"https://pith.science/pith/EGLX34FCUMWYMWTDRWTS3ITFVZ.json","view_paper":"https://pith.science/paper/EGLX34FC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.18540&json=true","fetch_graph":"https://pith.science/api/pith-number/EGLX34FCUMWYMWTDRWTS3ITFVZ/graph.json","fetch_events":"https://pith.science/api/pith-number/EGLX34FCUMWYMWTDRWTS3ITFVZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EGLX34FCUMWYMWTDRWTS3ITFVZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EGLX34FCUMWYMWTDRWTS3ITFVZ/action/storage_attestation","attest_author":"https://pith.science/pith/EGLX34FCUMWYMWTDRWTS3ITFVZ/action/author_attestation","sign_citation":"https://pith.science/pith/EGLX34FCUMWYMWTDRWTS3ITFVZ/action/citation_signature","submit_replication":"https://pith.science/pith/EGLX34FCUMWYMWTDRWTS3ITFVZ/action/replication_record"}},"created_at":"2026-07-05T10:21:19.568151+00:00","updated_at":"2026-07-05T10:21:19.568151+00:00"}