{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:JEH6X3OGKEODL65UJPA6C6SXLE","short_pith_number":"pith:JEH6X3OG","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"},"canonical_sha256":"490febedc6511c35fbb44bc1e17a5759012d13e4ce1d69c03b936df222602191","source":{"kind":"arxiv","id":"2412.02159","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2412.02159","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"arxiv_version","alias_value":"2412.02159v1","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.02159","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"pith_short_12","alias_value":"JEH6X3OGKEOD","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"pith_short_16","alias_value":"JEH6X3OGKEODL65U","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"pith_short_8","alias_value":"JEH6X3OG","created_at":"2026-07-05T09:43:45Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:JEH6X3OGKEODL65UJPA6C6SXLE","target":"record","payload":{"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"},"canonical_sha256":"490febedc6511c35fbb44bc1e17a5759012d13e4ce1d69c03b936df222602191","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"},"source_kind":"arxiv","source_id":"2412.02159","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T09:43:45Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"yRhXUqby7M09qCjjPsEHCZ3tEBG84QGwYuAqNPM3hEXTZhsKHBDdhSRLxxOyPcHV0GXFEhmiBB9IVqiKXGdDBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-14T10:09:44.561307Z"},"content_sha256":"bebbb088f64d02e970ba2abaf9db28efcea32341f30d10b6519f4654c536077d","schema_version":"1.0","event_id":"sha256:bebbb088f64d02e970ba2abaf9db28efcea32341f30d10b6519f4654c536077d"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:JEH6X3OGKEODL65UJPA6C6SXLE","target":"graph","payload":{"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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T09:43:45Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"DjVshs17KuQFDz6vkN1oqlo+yQaLoSo/+ctR/SC9BhHSSA//KVmYWmpdtKJlbv9h4nKoQEh0ctPHtYg76iWaCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-14T10:09:44.561830Z"},"content_sha256":"2d00d67590bd5b825a4a47a6986537474bd9f39545aa203cff55f17252dc8648","schema_version":"1.0","event_id":"sha256:2d00d67590bd5b825a4a47a6986537474bd9f39545aa203cff55f17252dc8648"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE/bundle.json","state_url":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/JEH6X3OGKEODL65UJPA6C6SXLE/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-14T10:09:44Z","links":{"resolver":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE","bundle":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE/bundle.json","state":"https://pith.science/pith/JEH6X3OGKEODL65UJPA6C6SXLE/state.json","well_known_bundle":"https://pith.science/.well-known/pith/JEH6X3OGKEODL65UJPA6C6SXLE/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:JEH6X3OGKEODL65UJPA6C6SXLE","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"6241e70c109decf037319a6c75461a3eb90fb223b157aba2e369a2939ea4e16d","cross_cats_sorted":["cs.AI","cs.CL","cs.CR"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-12-03T04:34:58Z","title_canon_sha256":"fe9828f4ce91a8f9c4a86ddf8f2951f30919e53ca18c1d681375ff09621524dc"},"schema_version":"1.0","source":{"id":"2412.02159","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2412.02159","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"arxiv_version","alias_value":"2412.02159v1","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.02159","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"pith_short_12","alias_value":"JEH6X3OGKEOD","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"pith_short_16","alias_value":"JEH6X3OGKEODL65U","created_at":"2026-07-05T09:43:45Z"},{"alias_kind":"pith_short_8","alias_value":"JEH6X3OG","created_at":"2026-07-05T09:43:45Z"}],"graph_snapshots":[{"event_id":"sha256:2d00d67590bd5b825a4a47a6986537474bd9f39545aa203cff55f17252dc8648","target":"graph","created_at":"2026-07-05T09:43:45Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2412.02159/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"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 ","authors_text":"Ethan Perez, Fazl Barez, Henry Sleight, Jesse Mu, John Hughes, Mrinank Sharma, Nir Shavit, Rajashree Agrawal, Rylan Schaeffer, Tony T. Wang","cross_cats":["cs.AI","cs.CL","cs.CR"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-12-03T04:34:58Z","title":"Jailbreak Defense in a Narrow Domain: Limitations of Existing Methods and a New Transcript-Classifier Approach"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.02159","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:bebbb088f64d02e970ba2abaf9db28efcea32341f30d10b6519f4654c536077d","target":"record","created_at":"2026-07-05T09:43:45Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"6241e70c109decf037319a6c75461a3eb90fb223b157aba2e369a2939ea4e16d","cross_cats_sorted":["cs.AI","cs.CL","cs.CR"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-12-03T04:34:58Z","title_canon_sha256":"fe9828f4ce91a8f9c4a86ddf8f2951f30919e53ca18c1d681375ff09621524dc"},"schema_version":"1.0","source":{"id":"2412.02159","kind":"arxiv","version":1}},"canonical_sha256":"490febedc6511c35fbb44bc1e17a5759012d13e4ce1d69c03b936df222602191","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"490febedc6511c35fbb44bc1e17a5759012d13e4ce1d69c03b936df222602191","first_computed_at":"2026-07-05T09:43:45.205148Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T09:43:45.205148Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"YdNk6OiUp4I3c7gPjdevmXQFAcE3ANbVx4lk4rd+YCYqftvcYAHWCV41B0/Rhw7+fXKmhUkzIDiayCdu6nUTBw==","signature_status":"signed_v1","signed_at":"2026-07-05T09:43:45.205587Z","signed_message":"canonical_sha256_bytes"},"source_id":"2412.02159","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:bebbb088f64d02e970ba2abaf9db28efcea32341f30d10b6519f4654c536077d","sha256:2d00d67590bd5b825a4a47a6986537474bd9f39545aa203cff55f17252dc8648"],"state_sha256":"2194c49a719f761e66cc6e639161c581ec0f83f7fffe490370fda0cf53014246"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"O97Wc+xB33cpwQa83Wo5xJD8Wl1gKfc8xD+ZpjW0xhHarrH1HzHsyxGgVmQro/8gSrgRZsISA+rqLjfdHyNcDg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-14T10:09:44.567606Z","bundle_sha256":"d0f362d2a559ba15a005d786fb598bd4e63ccf16bd8e9fd1b7c65e514b4ec02d"}}