{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:P7ZXU3BGC2QIWEVTIG7TH5H3M4","short_pith_number":"pith:P7ZXU3BG","schema_version":"1.0","canonical_sha256":"7ff37a6c2616a08b12b341bf33f4fb67205df8c26282b5d941eb47175a06f53a","source":{"kind":"arxiv","id":"2605.26772","version":1},"attestation_state":"computed","paper":{"title":"Beyond a Single Direction: Chain-of-Thought Disrupts Simple Steering of Refusal","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bela Gipp, Dominik Meier, Jiachen Zhao, Kia-J\\\"ung Yang, Terry Ruas","submitted_at":"2026-05-26T09:41:15Z","abstract_excerpt":"Large reasoning models (LRMs) generate chain-of-thought (CoT) traces before producing final outputs, introducing a dynamic internal state that may complicate control mechanisms such as refusal. Unlike instruction-tuned LLMs, where refusal is mediated by a single directional subspace, refusal in large reasoning models (LRMs) additionally depends on the CoT. In DeepSeek-R1-Distill-LLaMA-8B, activation steering reverses refusal in only 39% of cases when the CoT is kept fixed, but removing the CoT entirely increases this to 70%, indicating that the CoT actively reinforces refusal. In a two-stage i"},"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":"2605.26772","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-26T09:41:15Z","cross_cats_sorted":[],"title_canon_sha256":"0f113d3f5890f2fdd9e42a583de69dca32d260fcdfaf726a402dd5ee27d2f005","abstract_canon_sha256":"2de3a90ce968483a77569ff123d4f2f75304dc9847845ec144eab57295494623"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:06:11.875328Z","signature_b64":"JqY3xbibl6QBnbL9/zjYaczeyQnVI5KN8Xjh9jJby4zn1iDFht7t7wLW2t5W2NytFbtgmF/mKxxM6c4lat26Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7ff37a6c2616a08b12b341bf33f4fb67205df8c26282b5d941eb47175a06f53a","last_reissued_at":"2026-05-27T01:06:11.870565Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:06:11.870565Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond a Single Direction: Chain-of-Thought Disrupts Simple Steering of Refusal","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bela Gipp, Dominik Meier, Jiachen Zhao, Kia-J\\\"ung Yang, Terry Ruas","submitted_at":"2026-05-26T09:41:15Z","abstract_excerpt":"Large reasoning models (LRMs) generate chain-of-thought (CoT) traces before producing final outputs, introducing a dynamic internal state that may complicate control mechanisms such as refusal. Unlike instruction-tuned LLMs, where refusal is mediated by a single directional subspace, refusal in large reasoning models (LRMs) additionally depends on the CoT. In DeepSeek-R1-Distill-LLaMA-8B, activation steering reverses refusal in only 39% of cases when the CoT is kept fixed, but removing the CoT entirely increases this to 70%, indicating that the CoT actively reinforces refusal. In a two-stage i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.26772","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/2605.26772/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":"2605.26772","created_at":"2026-05-27T01:06:11.874521+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.26772v1","created_at":"2026-05-27T01:06:11.874521+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.26772","created_at":"2026-05-27T01:06:11.874521+00:00"},{"alias_kind":"pith_short_12","alias_value":"P7ZXU3BGC2QI","created_at":"2026-05-27T01:06:11.874521+00:00"},{"alias_kind":"pith_short_16","alias_value":"P7ZXU3BGC2QIWEVT","created_at":"2026-05-27T01:06:11.874521+00:00"},{"alias_kind":"pith_short_8","alias_value":"P7ZXU3BG","created_at":"2026-05-27T01:06:11.874521+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/P7ZXU3BGC2QIWEVTIG7TH5H3M4","json":"https://pith.science/pith/P7ZXU3BGC2QIWEVTIG7TH5H3M4.json","graph_json":"https://pith.science/api/pith-number/P7ZXU3BGC2QIWEVTIG7TH5H3M4/graph.json","events_json":"https://pith.science/api/pith-number/P7ZXU3BGC2QIWEVTIG7TH5H3M4/events.json","paper":"https://pith.science/paper/P7ZXU3BG"},"agent_actions":{"view_html":"https://pith.science/pith/P7ZXU3BGC2QIWEVTIG7TH5H3M4","download_json":"https://pith.science/pith/P7ZXU3BGC2QIWEVTIG7TH5H3M4.json","view_paper":"https://pith.science/paper/P7ZXU3BG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.26772&json=true","fetch_graph":"https://pith.science/api/pith-number/P7ZXU3BGC2QIWEVTIG7TH5H3M4/graph.json","fetch_events":"https://pith.science/api/pith-number/P7ZXU3BGC2QIWEVTIG7TH5H3M4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/P7ZXU3BGC2QIWEVTIG7TH5H3M4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/P7ZXU3BGC2QIWEVTIG7TH5H3M4/action/storage_attestation","attest_author":"https://pith.science/pith/P7ZXU3BGC2QIWEVTIG7TH5H3M4/action/author_attestation","sign_citation":"https://pith.science/pith/P7ZXU3BGC2QIWEVTIG7TH5H3M4/action/citation_signature","submit_replication":"https://pith.science/pith/P7ZXU3BGC2QIWEVTIG7TH5H3M4/action/replication_record"}},"created_at":"2026-05-27T01:06:11.874521+00:00","updated_at":"2026-05-27T01:06:11.874521+00:00"}