{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:JEH6X3OGKEODL65UJPA6C6SXLE","short_pith_number":"pith:JEH6X3OG","schema_version":"1.0","canonical_sha256":"490febedc6511c35fbb44bc1e17a5759012d13e4ce1d69c03b936df222602191","source":{"kind":"arxiv","id":"2412.02159","version":1},"attestation_state":"computed","paper":{"title":"Jailbreak Defense in a Narrow Domain: Limitations of Existing Methods and a New Transcript-Classifier Approach","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Ethan Perez, Fazl Barez, Henry Sleight, Jesse Mu, John Hughes, Mrinank Sharma, Nir Shavit, Rajashree Agrawal, Rylan Schaeffer, Tony T. Wang","submitted_at":"2024-12-03T04:34:58Z","abstract_excerpt":"Defending large language models against jailbreaks so that they never engage in a broadly-defined set of forbidden behaviors is an open problem. In this paper, we investigate the difficulty of jailbreak-defense when we only want to forbid a narrowly-defined set of behaviors. As a case study, we focus on preventing an LLM from helping a user make a bomb. We find that popular defenses such as safety training, adversarial training, and input/output classifiers are unable to fully solve this problem. In pursuit of a better solution, we develop a transcript-classifier defense which outperforms the "},"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":"2412.02159","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-12-03T04:34:58Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CR"],"title_canon_sha256":"fe9828f4ce91a8f9c4a86ddf8f2951f30919e53ca18c1d681375ff09621524dc","abstract_canon_sha256":"6241e70c109decf037319a6c75461a3eb90fb223b157aba2e369a2939ea4e16d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:43:45.205587Z","signature_b64":"YdNk6OiUp4I3c7gPjdevmXQFAcE3ANbVx4lk4rd+YCYqftvcYAHWCV41B0/Rhw7+fXKmhUkzIDiayCdu6nUTBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"490febedc6511c35fbb44bc1e17a5759012d13e4ce1d69c03b936df222602191","last_reissued_at":"2026-07-05T09:43:45.205148Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:43:45.205148Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Jailbreak Defense in a Narrow Domain: Limitations of Existing Methods and a New Transcript-Classifier Approach","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Ethan Perez, Fazl Barez, Henry Sleight, Jesse Mu, John Hughes, Mrinank Sharma, Nir Shavit, Rajashree Agrawal, Rylan Schaeffer, Tony T. Wang","submitted_at":"2024-12-03T04:34:58Z","abstract_excerpt":"Defending large language models against jailbreaks so that they never engage in a broadly-defined set of forbidden behaviors is an open problem. In this paper, we investigate the difficulty of jailbreak-defense when we only want to forbid a narrowly-defined set of behaviors. As a case study, we focus on preventing an LLM from helping a user make a bomb. We find that popular defenses such as safety training, adversarial training, and input/output classifiers are unable to fully solve this problem. In pursuit of a better solution, we develop a transcript-classifier defense which outperforms the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.02159","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/2412.02159/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":"2412.02159","created_at":"2026-07-05T09:43:45.205206+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.02159v1","created_at":"2026-07-05T09:43:45.205206+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.02159","created_at":"2026-07-05T09:43:45.205206+00:00"},{"alias_kind":"pith_short_12","alias_value":"JEH6X3OGKEOD","created_at":"2026-07-05T09:43:45.205206+00:00"},{"alias_kind":"pith_short_16","alias_value":"JEH6X3OGKEODL65U","created_at":"2026-07-05T09:43:45.205206+00:00"},{"alias_kind":"pith_short_8","alias_value":"JEH6X3OG","created_at":"2026-07-05T09:43:45.205206+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE","json":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE.json","graph_json":"https://pith.science/api/pith-number/JEH6X3OGKEODL65UJPA6C6SXLE/graph.json","events_json":"https://pith.science/api/pith-number/JEH6X3OGKEODL65UJPA6C6SXLE/events.json","paper":"https://pith.science/paper/JEH6X3OG"},"agent_actions":{"view_html":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE","download_json":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE.json","view_paper":"https://pith.science/paper/JEH6X3OG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.02159&json=true","fetch_graph":"https://pith.science/api/pith-number/JEH6X3OGKEODL65UJPA6C6SXLE/graph.json","fetch_events":"https://pith.science/api/pith-number/JEH6X3OGKEODL65UJPA6C6SXLE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE/action/storage_attestation","attest_author":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE/action/author_attestation","sign_citation":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE/action/citation_signature","submit_replication":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE/action/replication_record"}},"created_at":"2026-07-05T09:43:45.205206+00:00","updated_at":"2026-07-05T09:43:45.205206+00:00"}