{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:FOFMCNFAVIODTO66EDJAV3FLFT","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":"c405e9fefa081d3075cb7b8ebab9f709732a1e9d1933e1ff895e39289b8ba040","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-02-26T10:03:33Z","title_canon_sha256":"839e602aeb035f0f037d466258d8506eacd129c0211e90a3e53fceeccab75675"},"schema_version":"1.0","source":{"id":"2402.16459","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2402.16459","created_at":"2026-07-05T08:28:38Z"},{"alias_kind":"arxiv_version","alias_value":"2402.16459v3","created_at":"2026-07-05T08:28:38Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.16459","created_at":"2026-07-05T08:28:38Z"},{"alias_kind":"pith_short_12","alias_value":"FOFMCNFAVIOD","created_at":"2026-07-05T08:28:38Z"},{"alias_kind":"pith_short_16","alias_value":"FOFMCNFAVIODTO66","created_at":"2026-07-05T08:28:38Z"},{"alias_kind":"pith_short_8","alias_value":"FOFMCNFA","created_at":"2026-07-05T08:28:38Z"}],"graph_snapshots":[{"event_id":"sha256:a2754bae0fea73fd1c3807d30c3a991a9f76f419179de0b9ed27168776061b44","target":"graph","created_at":"2026-07-05T08:28:38Z","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/2402.16459/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Although many large language models (LLMs) have been trained to refuse harmful requests, they are still vulnerable to jailbreaking attacks which rewrite the original prompt to conceal its harmful intent. In this paper, we propose a new method for defending LLMs against jailbreaking attacks by ``backtranslation''. Specifically, given an initial response generated by the target LLM from an input prompt, our backtranslation prompts a language model to infer an input prompt that can lead to the response. The inferred prompt is called the backtranslated prompt which tends to reveal the actual inten","authors_text":"Andrew Bai, Cho-Jui Hsieh, Yihan Wang, Zhouxing Shi","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-02-26T10:03:33Z","title":"Defending LLMs against Jailbreaking Attacks via Backtranslation"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.16459","kind":"arxiv","version":3},"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":"sha256:1bf529c95925c40d1ecac271ba16f01bda5269741fa626725222ac0cdc582378","target":"record","created_at":"2026-07-05T08:28:38Z","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":"c405e9fefa081d3075cb7b8ebab9f709732a1e9d1933e1ff895e39289b8ba040","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-02-26T10:03:33Z","title_canon_sha256":"839e602aeb035f0f037d466258d8506eacd129c0211e90a3e53fceeccab75675"},"schema_version":"1.0","source":{"id":"2402.16459","kind":"arxiv","version":3}},"canonical_sha256":"2b8ac134a0aa1c39bbde20d20aecab2cc7270a2ebedb94c5a82892b0dff724e0","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"2b8ac134a0aa1c39bbde20d20aecab2cc7270a2ebedb94c5a82892b0dff724e0","first_computed_at":"2026-07-05T08:28:38.944788Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T08:28:38.944788Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"qQMJPv0b0wsYXCm84mCjneaiBcpYYcPsaMgONIFKP3ldhVmtAzO+z1p1nA08KuQNmC4FPyg0YInaHfKE09nhCg==","signature_status":"signed_v1","signed_at":"2026-07-05T08:28:38.945252Z","signed_message":"canonical_sha256_bytes"},"source_id":"2402.16459","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:1bf529c95925c40d1ecac271ba16f01bda5269741fa626725222ac0cdc582378","sha256:a2754bae0fea73fd1c3807d30c3a991a9f76f419179de0b9ed27168776061b44"],"state_sha256":"b8610242be551c4309add766f3b03f83750b06c8b83222f8590ccc80ec481814"}