{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Q5HZRNCCBR23VZVFCSQNJLVJIR","short_pith_number":"pith:Q5HZRNCC","schema_version":"1.0","canonical_sha256":"874f98b4420c75bae6a514a0d4aea94453712bbad5e6ffa4741dfbdf1caea8e9","source":{"kind":"arxiv","id":"2505.18807","version":1},"attestation_state":"computed","paper":{"title":"Mitigating Deceptive Alignment via Self-Monitoring","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Boyuan Chen, Donghai Hong, Jiaming Ji, Jiayi Zhou, Juntao Dai, Kaile Wang, Sirui Han, Sitong Fang, Wenqi Chen, Yaodong Yang, Yike Guo","submitted_at":"2025-05-24T17:41:47Z","abstract_excerpt":"Modern large language models rely on chain-of-thought (CoT) reasoning to achieve impressive performance, yet the same mechanism can amplify deceptive alignment, situations in which a model appears aligned while covertly pursuing misaligned goals. Existing safety pipelines treat deception as a black-box output to be filtered post-hoc, leaving the model free to scheme during its internal reasoning. We ask: Can deception be intercepted while the model is thinking? We answer this question, the first framework that embeds a Self-Monitor inside the CoT process itself, named CoT Monitor+. During gene"},"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":"2505.18807","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-05-24T17:41:47Z","cross_cats_sorted":[],"title_canon_sha256":"1e25620bee427f14aff080f8cbf4b9cdc9edd0a6296589341cf05ea179c0fbb4","abstract_canon_sha256":"5acab73228edf748b92442cbd2e84cbac7ee0044354bc7be16ad552fc4d4e4c2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:19.951599Z","signature_b64":"+6IAT+Y3Zrq4etiKEJSBIV5+zdpsZywX6FEBrXS1LekMGiM9U5ymzGW+1djIf4egZruSutPE0ch0iow+kJlSDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"874f98b4420c75bae6a514a0d4aea94453712bbad5e6ffa4741dfbdf1caea8e9","last_reissued_at":"2026-07-05T11:09:19.951052Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:19.951052Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mitigating Deceptive Alignment via Self-Monitoring","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Boyuan Chen, Donghai Hong, Jiaming Ji, Jiayi Zhou, Juntao Dai, Kaile Wang, Sirui Han, Sitong Fang, Wenqi Chen, Yaodong Yang, Yike Guo","submitted_at":"2025-05-24T17:41:47Z","abstract_excerpt":"Modern large language models rely on chain-of-thought (CoT) reasoning to achieve impressive performance, yet the same mechanism can amplify deceptive alignment, situations in which a model appears aligned while covertly pursuing misaligned goals. Existing safety pipelines treat deception as a black-box output to be filtered post-hoc, leaving the model free to scheme during its internal reasoning. We ask: Can deception be intercepted while the model is thinking? We answer this question, the first framework that embeds a Self-Monitor inside the CoT process itself, named CoT Monitor+. During gene"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.18807","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/2505.18807/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":"2505.18807","created_at":"2026-07-05T11:09:19.951121+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.18807v1","created_at":"2026-07-05T11:09:19.951121+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.18807","created_at":"2026-07-05T11:09:19.951121+00:00"},{"alias_kind":"pith_short_12","alias_value":"Q5HZRNCCBR23","created_at":"2026-07-05T11:09:19.951121+00:00"},{"alias_kind":"pith_short_16","alias_value":"Q5HZRNCCBR23VZVF","created_at":"2026-07-05T11:09:19.951121+00:00"},{"alias_kind":"pith_short_8","alias_value":"Q5HZRNCC","created_at":"2026-07-05T11:09:19.951121+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25899","citing_title":"Manipulation Is Task-Dependent: A Multi-Axis, Multi-Environment Evaluation of Frontier LLMs","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30169","citing_title":"Dissociative Identity: Language Model Agents Lack Grounding for Reputation Mechanisms","ref_index":67,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Q5HZRNCCBR23VZVFCSQNJLVJIR","json":"https://pith.science/pith/Q5HZRNCCBR23VZVFCSQNJLVJIR.json","graph_json":"https://pith.science/api/pith-number/Q5HZRNCCBR23VZVFCSQNJLVJIR/graph.json","events_json":"https://pith.science/api/pith-number/Q5HZRNCCBR23VZVFCSQNJLVJIR/events.json","paper":"https://pith.science/paper/Q5HZRNCC"},"agent_actions":{"view_html":"https://pith.science/pith/Q5HZRNCCBR23VZVFCSQNJLVJIR","download_json":"https://pith.science/pith/Q5HZRNCCBR23VZVFCSQNJLVJIR.json","view_paper":"https://pith.science/paper/Q5HZRNCC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.18807&json=true","fetch_graph":"https://pith.science/api/pith-number/Q5HZRNCCBR23VZVFCSQNJLVJIR/graph.json","fetch_events":"https://pith.science/api/pith-number/Q5HZRNCCBR23VZVFCSQNJLVJIR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Q5HZRNCCBR23VZVFCSQNJLVJIR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Q5HZRNCCBR23VZVFCSQNJLVJIR/action/storage_attestation","attest_author":"https://pith.science/pith/Q5HZRNCCBR23VZVFCSQNJLVJIR/action/author_attestation","sign_citation":"https://pith.science/pith/Q5HZRNCCBR23VZVFCSQNJLVJIR/action/citation_signature","submit_replication":"https://pith.science/pith/Q5HZRNCCBR23VZVFCSQNJLVJIR/action/replication_record"}},"created_at":"2026-07-05T11:09:19.951121+00:00","updated_at":"2026-07-05T11:09:19.951121+00:00"}