{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GYNNPMA5QXKJ6W4B2XDTX7RLZL","short_pith_number":"pith:GYNNPMA5","schema_version":"1.0","canonical_sha256":"361ad7b01d85d49f5b81d5c73bfe2bcad339e7ab37ed5aec141319806b9b30b6","source":{"kind":"arxiv","id":"2503.19326","version":2},"attestation_state":"computed","paper":{"title":"Process or Result? Manipulated Ending Tokens Can Mislead Reasoning LLMs to Ignore the Correct Reasoning Steps","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bryan Hooi, Yiwei Wang, Yu Cui, Yujun Cai","submitted_at":"2025-03-25T03:43:11Z","abstract_excerpt":"Recent reasoning large language models (LLMs) have demonstrated remarkable improvements in mathematical reasoning capabilities through long Chain-of-Thought. The reasoning tokens of these models enable self-correction within reasoning chains, enhancing robustness. This motivates our exploration: how vulnerable are reasoning LLMs to subtle errors in their input reasoning chains? We introduce \"Compromising Thought\" (CPT), a vulnerability where models presented with reasoning tokens containing manipulated calculation results tend to ignore correct reasoning steps and adopt incorrect results inste"},"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":"2503.19326","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-03-25T03:43:11Z","cross_cats_sorted":[],"title_canon_sha256":"93e9254ba9feb7112f67700711bbecf1148dc88690c61d0907098d800420c8a2","abstract_canon_sha256":"9999bf754a916533d22c96fad1184b9915df2fdffccb852a2967e7592930d3b2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:42:16.063530Z","signature_b64":"sK2KOmtLY/QQVsRGYDqdm3ewqmdlIcbk+D9VJit+Rqlx9azXS9XNd7mO0jsAr52GDNE2o3nwxjU8aLkDlb+oCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"361ad7b01d85d49f5b81d5c73bfe2bcad339e7ab37ed5aec141319806b9b30b6","last_reissued_at":"2026-07-05T10:42:16.063049Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:42:16.063049Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Process or Result? Manipulated Ending Tokens Can Mislead Reasoning LLMs to Ignore the Correct Reasoning Steps","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bryan Hooi, Yiwei Wang, Yu Cui, Yujun Cai","submitted_at":"2025-03-25T03:43:11Z","abstract_excerpt":"Recent reasoning large language models (LLMs) have demonstrated remarkable improvements in mathematical reasoning capabilities through long Chain-of-Thought. The reasoning tokens of these models enable self-correction within reasoning chains, enhancing robustness. This motivates our exploration: how vulnerable are reasoning LLMs to subtle errors in their input reasoning chains? We introduce \"Compromising Thought\" (CPT), a vulnerability where models presented with reasoning tokens containing manipulated calculation results tend to ignore correct reasoning steps and adopt incorrect results inste"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.19326","kind":"arxiv","version":2},"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/2503.19326/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":"2503.19326","created_at":"2026-07-05T10:42:16.063108+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.19326v2","created_at":"2026-07-05T10:42:16.063108+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.19326","created_at":"2026-07-05T10:42:16.063108+00:00"},{"alias_kind":"pith_short_12","alias_value":"GYNNPMA5QXKJ","created_at":"2026-07-05T10:42:16.063108+00:00"},{"alias_kind":"pith_short_16","alias_value":"GYNNPMA5QXKJ6W4B","created_at":"2026-07-05T10:42:16.063108+00:00"},{"alias_kind":"pith_short_8","alias_value":"GYNNPMA5","created_at":"2026-07-05T10:42:16.063108+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2503.09567","citing_title":"Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models","ref_index":150,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GYNNPMA5QXKJ6W4B2XDTX7RLZL","json":"https://pith.science/pith/GYNNPMA5QXKJ6W4B2XDTX7RLZL.json","graph_json":"https://pith.science/api/pith-number/GYNNPMA5QXKJ6W4B2XDTX7RLZL/graph.json","events_json":"https://pith.science/api/pith-number/GYNNPMA5QXKJ6W4B2XDTX7RLZL/events.json","paper":"https://pith.science/paper/GYNNPMA5"},"agent_actions":{"view_html":"https://pith.science/pith/GYNNPMA5QXKJ6W4B2XDTX7RLZL","download_json":"https://pith.science/pith/GYNNPMA5QXKJ6W4B2XDTX7RLZL.json","view_paper":"https://pith.science/paper/GYNNPMA5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.19326&json=true","fetch_graph":"https://pith.science/api/pith-number/GYNNPMA5QXKJ6W4B2XDTX7RLZL/graph.json","fetch_events":"https://pith.science/api/pith-number/GYNNPMA5QXKJ6W4B2XDTX7RLZL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GYNNPMA5QXKJ6W4B2XDTX7RLZL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GYNNPMA5QXKJ6W4B2XDTX7RLZL/action/storage_attestation","attest_author":"https://pith.science/pith/GYNNPMA5QXKJ6W4B2XDTX7RLZL/action/author_attestation","sign_citation":"https://pith.science/pith/GYNNPMA5QXKJ6W4B2XDTX7RLZL/action/citation_signature","submit_replication":"https://pith.science/pith/GYNNPMA5QXKJ6W4B2XDTX7RLZL/action/replication_record"}},"created_at":"2026-07-05T10:42:16.063108+00:00","updated_at":"2026-07-05T10:42:16.063108+00:00"}