{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ARMCGDQE2V5KDGL7FQQZJCWMYG","short_pith_number":"pith:ARMCGDQE","schema_version":"1.0","canonical_sha256":"0458230e04d57aa1997f2c21948accc1965dbeb706e9499129f15696d1269661","source":{"kind":"arxiv","id":"2510.26418","version":4},"attestation_state":"computed","paper":{"title":"Chain-of-Thought Hijacking","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Fazl Barez, Jianli Zhao, Mrinank Sharma, Rylan Schaeffer, Tingchen Fu","submitted_at":"2025-10-30T12:10:03Z","abstract_excerpt":"Large Reasoning Models (LRMs) improve task performance through extended inference-time reasoning. Although previous studies suggest that longer reasoning should lead to more robust safety behavior, we find evidence to the contrary: over-extended reasoning can instead be exploited to systematically weaken refusal behavior. We propose Chain-of-Thought Hijacking, a simple yet effective black-box jailbreak attack that induces LRMs to engage in prolonged benign puzzle-solving reasoning, often lasting more than five minutes, before eliciting harmful compliance. Across HarmBench, CoT Hijacking achiev"},"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":"2510.26418","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-10-30T12:10:03Z","cross_cats_sorted":[],"title_canon_sha256":"52e93352d29bc904afed2c494aa8028596e71729c2e9167cf9fef3e50293d3b9","abstract_canon_sha256":"5795eb1cda8c62ec4b3e15395dc3f2cccfeb4635e5153e1c1d5f25cc44c6d041"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:03:18.314117Z","signature_b64":"NQid23hMHAnrBqQfHR+ILgNrkaYLV+OjSj5UVkEuMBkFQoq+VfVNj28X0/iEwl4jH7Vj0y3v8kjRzpnQvOo7BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0458230e04d57aa1997f2c21948accc1965dbeb706e9499129f15696d1269661","last_reissued_at":"2026-05-26T01:03:18.313275Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:03:18.313275Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Chain-of-Thought Hijacking","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Fazl Barez, Jianli Zhao, Mrinank Sharma, Rylan Schaeffer, Tingchen Fu","submitted_at":"2025-10-30T12:10:03Z","abstract_excerpt":"Large Reasoning Models (LRMs) improve task performance through extended inference-time reasoning. Although previous studies suggest that longer reasoning should lead to more robust safety behavior, we find evidence to the contrary: over-extended reasoning can instead be exploited to systematically weaken refusal behavior. We propose Chain-of-Thought Hijacking, a simple yet effective black-box jailbreak attack that induces LRMs to engage in prolonged benign puzzle-solving reasoning, often lasting more than five minutes, before eliciting harmful compliance. Across HarmBench, CoT Hijacking achiev"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.26418","kind":"arxiv","version":4},"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/2510.26418/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":"2510.26418","created_at":"2026-05-26T01:03:18.313392+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.26418v4","created_at":"2026-05-26T01:03:18.313392+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.26418","created_at":"2026-05-26T01:03:18.313392+00:00"},{"alias_kind":"pith_short_12","alias_value":"ARMCGDQE2V5K","created_at":"2026-05-26T01:03:18.313392+00:00"},{"alias_kind":"pith_short_16","alias_value":"ARMCGDQE2V5KDGL7","created_at":"2026-05-26T01:03:18.313392+00:00"},{"alias_kind":"pith_short_8","alias_value":"ARMCGDQE","created_at":"2026-05-26T01:03:18.313392+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2605.19485","citing_title":"Attention-Guided Reward for Reinforcement Learning-based Jailbreak against Large Reasoning Models","ref_index":40,"is_internal_anchor":true},{"citing_arxiv_id":"2603.09002","citing_title":"Security Considerations for Multi-agent Systems","ref_index":143,"is_internal_anchor":true},{"citing_arxiv_id":"2603.17368","citing_title":"Towards Safer Large Reasoning Models by Promoting Safety Decision-Making before Chain-of-Thought Generation","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09235","citing_title":"Unreal Thinking: Chain-of-Thought Hijacking via Two-stage Backdoor","ref_index":4,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ARMCGDQE2V5KDGL7FQQZJCWMYG","json":"https://pith.science/pith/ARMCGDQE2V5KDGL7FQQZJCWMYG.json","graph_json":"https://pith.science/api/pith-number/ARMCGDQE2V5KDGL7FQQZJCWMYG/graph.json","events_json":"https://pith.science/api/pith-number/ARMCGDQE2V5KDGL7FQQZJCWMYG/events.json","paper":"https://pith.science/paper/ARMCGDQE"},"agent_actions":{"view_html":"https://pith.science/pith/ARMCGDQE2V5KDGL7FQQZJCWMYG","download_json":"https://pith.science/pith/ARMCGDQE2V5KDGL7FQQZJCWMYG.json","view_paper":"https://pith.science/paper/ARMCGDQE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.26418&json=true","fetch_graph":"https://pith.science/api/pith-number/ARMCGDQE2V5KDGL7FQQZJCWMYG/graph.json","fetch_events":"https://pith.science/api/pith-number/ARMCGDQE2V5KDGL7FQQZJCWMYG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ARMCGDQE2V5KDGL7FQQZJCWMYG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ARMCGDQE2V5KDGL7FQQZJCWMYG/action/storage_attestation","attest_author":"https://pith.science/pith/ARMCGDQE2V5KDGL7FQQZJCWMYG/action/author_attestation","sign_citation":"https://pith.science/pith/ARMCGDQE2V5KDGL7FQQZJCWMYG/action/citation_signature","submit_replication":"https://pith.science/pith/ARMCGDQE2V5KDGL7FQQZJCWMYG/action/replication_record"}},"created_at":"2026-05-26T01:03:18.313392+00:00","updated_at":"2026-05-26T01:03:18.313392+00:00"}