{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:HPXY4FXAWCTEN2H7PBE2R2XC4Q","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":"4d5e7e816336e929d7d40b745bf87d1f01056f4b4e36a84ac78b01c0d48be3e8","cross_cats_sorted":["cs.AI","cs.CR"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-10-10T20:15:54Z","title_canon_sha256":"5f186a839f5029fedc22e2a80147814d1871c88212d9f86a9eacb53073f5763e"},"schema_version":"1.0","source":{"id":"2310.06987","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2310.06987","created_at":"2026-05-17T23:38:46Z"},{"alias_kind":"arxiv_version","alias_value":"2310.06987v1","created_at":"2026-05-17T23:38:46Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.06987","created_at":"2026-05-17T23:38:46Z"},{"alias_kind":"pith_short_12","alias_value":"HPXY4FXAWCTE","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"HPXY4FXAWCTEN2H7","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"HPXY4FXA","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:af437a6c0098117c0d9867dd70a9bfbb302e27d8f6791f812580718da9fdda16","target":"graph","created_at":"2026-05-17T23:38:46Z","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":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"By exploiting different generation strategies, including varying decoding hyper-parameters and sampling methods, we increase the misalignment rate from 0% to more than 95% across 11 language models including LLaMA2, Vicuna, Falcon, and MPT families, outperforming state-of-the-art attacks with 30× lower computational cost."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the high misalignment rates result specifically from the generation exploitation rather than from the choice of test prompts or from model-specific quirks that would not generalize to other prompts or models."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Varying decoding strategies such as temperature and sampling methods jailbreaks safety alignments in open-source LLMs, raising misalignment from 0% to over 95% at 30x lower cost than prior attacks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Varying decoding parameters and sampling methods jailbreak aligned open-source LLMs, raising misalignment from 0% to over 95%."}],"snapshot_sha256":"dfd0c8a92b32a145c080c9241b311f3747d914ea5dbb902a1704056c046babd4"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2c7f1c8707dbad0eea3616ee40e041d16fd183337d54ed16bff418a7aabf9c9a"},"paper":{"abstract_excerpt":"The rapid progress in open-source large language models (LLMs) is significantly advancing AI development. Extensive efforts have been made before model release to align their behavior with human values, with the primary goal of ensuring their helpfulness and harmlessness. However, even carefully aligned models can be manipulated maliciously, leading to unintended behaviors, known as \"jailbreaks\". These jailbreaks are typically triggered by specific text inputs, often referred to as adversarial prompts. In this work, we propose the generation exploitation attack, an extremely simple approach th","authors_text":"Danqi Chen, Kai Li, Mengzhou Xia, Samyak Gupta, Yangsibo Huang","cross_cats":["cs.AI","cs.CR"],"headline":"Varying decoding parameters and sampling methods jailbreak aligned open-source LLMs, raising misalignment from 0% to over 95%.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-10-10T20:15:54Z","title":"Catastrophic Jailbreak of Open-source LLMs via Exploiting Generation"},"references":{"count":25,"internal_anchors":12,"resolved_work":25,"sample":[{"cited_arxiv_id":"2305.10403","doi":"","is_internal_anchor":true,"ref_index":1,"title":"PaLM 2 Technical Report","work_id":"905ee9a7-ea61-4a94-bd62-2600cbe3e315","year":null},{"cited_arxiv_id":"2112.00861","doi":"","is_internal_anchor":true,"ref_index":2,"title":"A General Language Assistant as a Laboratory for Alignment","work_id":"a43f9ea0-01be-47d5-b8ee-a1a9f73381c5","year":null},{"cited_arxiv_id":"2204.05862","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback","work_id":"a1f2574b-a899-4713-be60-c87ba332656c","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Choquette-Choo , Matthew Jagielski, Irena Gao, Anas Awadalla, Pang Wei Koh, Daphne Ippolito, Katherine Lee, Florian Tramer, and Ludwig Schmidt","work_id":"cbf9eae3-40c5-419a-9f5b-335eab5a2bb2","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Explore, Establish , Exploit : Red Teaming Language Models from Scratch","work_id":"c346acec-8e7c-4c8c-9091-3ad142e7a978","year":null}],"snapshot_sha256":"cfa6d1c64d262bf5e7e48da8d9dc365c907687f71cc5387b7159af5471373897"},"source":{"id":"2310.06987","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-16T21:56:16.545421Z","id":"4d58dadc-9a8e-4950-a763-4202d91fda83","model_set":{"reader":"grok-4.3"},"one_line_summary":"Varying decoding strategies such as temperature and sampling methods jailbreaks safety alignments in open-source LLMs, raising misalignment from 0% to over 95% at 30x lower cost than prior attacks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Varying decoding parameters and sampling methods jailbreak aligned open-source LLMs, raising misalignment from 0% to over 95%.","strongest_claim":"By exploiting different generation strategies, including varying decoding hyper-parameters and sampling methods, we increase the misalignment rate from 0% to more than 95% across 11 language models including LLaMA2, Vicuna, Falcon, and MPT families, outperforming state-of-the-art attacks with 30× lower computational cost.","weakest_assumption":"That the high misalignment rates result specifically from the generation exploitation rather than from the choice of test prompts or from model-specific quirks that would not generalize to other prompts or models."}},"verdict_id":"4d58dadc-9a8e-4950-a763-4202d91fda83"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:22d8299873f835d6c65fad195c128b050731b58cb502b2ff29883b4009a70845","target":"record","created_at":"2026-05-17T23:38:46Z","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":"4d5e7e816336e929d7d40b745bf87d1f01056f4b4e36a84ac78b01c0d48be3e8","cross_cats_sorted":["cs.AI","cs.CR"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-10-10T20:15:54Z","title_canon_sha256":"5f186a839f5029fedc22e2a80147814d1871c88212d9f86a9eacb53073f5763e"},"schema_version":"1.0","source":{"id":"2310.06987","kind":"arxiv","version":1}},"canonical_sha256":"3bef8e16e0b0a646e8ff7849a8eae2e40a79ae77c5a54543348eff8abbcd663e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"3bef8e16e0b0a646e8ff7849a8eae2e40a79ae77c5a54543348eff8abbcd663e","first_computed_at":"2026-05-17T23:38:46.442598Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:46.442598Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"sIJJtl82eSBFWHgGzB8bBqAgQKHooEJmBlZ43rznB/uv77GfcTEp7Q7fTCpQt7D3e3jqXKvhHNfu2/4P5jMzCg==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:46.443047Z","signed_message":"canonical_sha256_bytes"},"source_id":"2310.06987","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:22d8299873f835d6c65fad195c128b050731b58cb502b2ff29883b4009a70845","sha256:af437a6c0098117c0d9867dd70a9bfbb302e27d8f6791f812580718da9fdda16"],"state_sha256":"6fc6b372a5f11571455b8eaf90442d12471197a218a7af30884d69383ffd455b"}