{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:2IRUGJIMQFFHEVKS4XO67O3JVA","short_pith_number":"pith:2IRUGJIM","schema_version":"1.0","canonical_sha256":"d22343250c814a725552e5ddefbb69a80dc6b909167ac6c2339bd1c821d3217f","source":{"kind":"arxiv","id":"2410.01294","version":3},"attestation_state":"computed","paper":{"title":"Endless Jailbreaks with Bijection Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Brian R.Y. Huang, Leonard Tang, Maximilian Li","submitted_at":"2024-10-02T07:40:56Z","abstract_excerpt":"Despite extensive safety measures, LLMs are vulnerable to adversarial inputs, or jailbreaks, which can elicit unsafe behaviors. In this work, we introduce bijection learning, a powerful attack algorithm which automatically fuzzes LLMs for safety vulnerabilities using randomly-generated encodings whose complexity can be tightly controlled. We leverage in-context learning to teach models bijective encodings, pass encoded queries to the model to bypass built-in safety mechanisms, and finally decode responses back into English. Our attack is extremely effective on a wide range of frontier language"},"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":"2410.01294","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-02T07:40:56Z","cross_cats_sorted":[],"title_canon_sha256":"2881475bc8f58c043538b7e2099820d4f5bd030c54e9fc4fff7ef93adbda2c66","abstract_canon_sha256":"e944ce90ed7a48372db694708a48e6d58d9397cbded33507e698b45173a41cdc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:01:01.847445Z","signature_b64":"uUOo9gPnL72yQPLUzjqdYSrqBzrO9Oe8wTzzfLpy/PycCwgdx28scSEuAfA1VPR9WVSRKOijqKr+9cL1EzWeCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d22343250c814a725552e5ddefbb69a80dc6b909167ac6c2339bd1c821d3217f","last_reissued_at":"2026-07-05T11:01:01.846974Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:01:01.846974Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Endless Jailbreaks with Bijection Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Brian R.Y. Huang, Leonard Tang, Maximilian Li","submitted_at":"2024-10-02T07:40:56Z","abstract_excerpt":"Despite extensive safety measures, LLMs are vulnerable to adversarial inputs, or jailbreaks, which can elicit unsafe behaviors. In this work, we introduce bijection learning, a powerful attack algorithm which automatically fuzzes LLMs for safety vulnerabilities using randomly-generated encodings whose complexity can be tightly controlled. We leverage in-context learning to teach models bijective encodings, pass encoded queries to the model to bypass built-in safety mechanisms, and finally decode responses back into English. Our attack is extremely effective on a wide range of frontier language"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.01294","kind":"arxiv","version":3},"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/2410.01294/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":"2410.01294","created_at":"2026-07-05T11:01:01.847030+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.01294v3","created_at":"2026-07-05T11:01:01.847030+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.01294","created_at":"2026-07-05T11:01:01.847030+00:00"},{"alias_kind":"pith_short_12","alias_value":"2IRUGJIMQFFH","created_at":"2026-07-05T11:01:01.847030+00:00"},{"alias_kind":"pith_short_16","alias_value":"2IRUGJIMQFFHEVKS","created_at":"2026-07-05T11:01:01.847030+00:00"},{"alias_kind":"pith_short_8","alias_value":"2IRUGJIM","created_at":"2026-07-05T11:01:01.847030+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2511.02356","citing_title":"ASTRA: An Automated Framework for Strategy Discovery, Retrieval, and Evolution for Jailbreaking LLMs","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09070","citing_title":"Single-Configuration Attack Success Rate Is Not Enough: Jailbreak Evaluations Should Report Distributional Attack Success","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2IRUGJIMQFFHEVKS4XO67O3JVA","json":"https://pith.science/pith/2IRUGJIMQFFHEVKS4XO67O3JVA.json","graph_json":"https://pith.science/api/pith-number/2IRUGJIMQFFHEVKS4XO67O3JVA/graph.json","events_json":"https://pith.science/api/pith-number/2IRUGJIMQFFHEVKS4XO67O3JVA/events.json","paper":"https://pith.science/paper/2IRUGJIM"},"agent_actions":{"view_html":"https://pith.science/pith/2IRUGJIMQFFHEVKS4XO67O3JVA","download_json":"https://pith.science/pith/2IRUGJIMQFFHEVKS4XO67O3JVA.json","view_paper":"https://pith.science/paper/2IRUGJIM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.01294&json=true","fetch_graph":"https://pith.science/api/pith-number/2IRUGJIMQFFHEVKS4XO67O3JVA/graph.json","fetch_events":"https://pith.science/api/pith-number/2IRUGJIMQFFHEVKS4XO67O3JVA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2IRUGJIMQFFHEVKS4XO67O3JVA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2IRUGJIMQFFHEVKS4XO67O3JVA/action/storage_attestation","attest_author":"https://pith.science/pith/2IRUGJIMQFFHEVKS4XO67O3JVA/action/author_attestation","sign_citation":"https://pith.science/pith/2IRUGJIMQFFHEVKS4XO67O3JVA/action/citation_signature","submit_replication":"https://pith.science/pith/2IRUGJIMQFFHEVKS4XO67O3JVA/action/replication_record"}},"created_at":"2026-07-05T11:01:01.847030+00:00","updated_at":"2026-07-05T11:01:01.847030+00:00"}