{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:K4QC7AD37YPFYKXY4NXGJM5SJA","short_pith_number":"pith:K4QC7AD3","schema_version":"1.0","canonical_sha256":"57202f807bfe1e5c2af8e36e64b3b248080166f1d2b43f5d1e39b7547790657d","source":{"kind":"arxiv","id":"2607.09786","version":1},"attestation_state":"computed","paper":{"title":"Length Penalties Make Chain-of-Thought Less Monitorable","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Bryce Little","submitted_at":"2026-07-08T14:18:26Z","abstract_excerpt":"Length-penalized reinforcement learning can shorten chain-of-thought reasoning while hiding an influence that drives the model's answer. In our experiments, training with length penalties does not stop misleading hints from steering models, even though the models' chains of thought mention the hint much less often. A token-accuracy evaluation would count these runs as successful because they use fewer reasoning tokens with little accuracy loss; it would miss whether the remaining trace still shows what drove the answer. We train Qwen3-4B and Qwen3-14B variants with different target chain lengt"},"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":"2607.09786","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-07-08T14:18:26Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"09cd21e1830a144a1aeb25c5948bde91d336affff8267ded711c6c48dea6f53b","abstract_canon_sha256":"f66005ade48ca68546466db9c26c2af851025c594e6f64841ee38f13a12ed859"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-14T00:18:35.697939Z","signature_b64":"Q7bYp/ULFfE55oWfFmEHxFmqPUl03eRMk9l3kquXHYUumUS+4CtYVHDHCtBjco/9Rh47xC/P1FKGka3x3P8yBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"57202f807bfe1e5c2af8e36e64b3b248080166f1d2b43f5d1e39b7547790657d","last_reissued_at":"2026-07-14T00:18:35.697036Z","signature_status":"signed_v1","first_computed_at":"2026-07-14T00:18:35.697036Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Length Penalties Make Chain-of-Thought Less Monitorable","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Bryce Little","submitted_at":"2026-07-08T14:18:26Z","abstract_excerpt":"Length-penalized reinforcement learning can shorten chain-of-thought reasoning while hiding an influence that drives the model's answer. In our experiments, training with length penalties does not stop misleading hints from steering models, even though the models' chains of thought mention the hint much less often. A token-accuracy evaluation would count these runs as successful because they use fewer reasoning tokens with little accuracy loss; it would miss whether the remaining trace still shows what drove the answer. We train Qwen3-4B and Qwen3-14B variants with different target chain lengt"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.09786","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/2607.09786/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":"2607.09786","created_at":"2026-07-14T00:18:35.697502+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.09786v1","created_at":"2026-07-14T00:18:35.697502+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.09786","created_at":"2026-07-14T00:18:35.697502+00:00"},{"alias_kind":"pith_short_12","alias_value":"K4QC7AD37YPF","created_at":"2026-07-14T00:18:35.697502+00:00"},{"alias_kind":"pith_short_16","alias_value":"K4QC7AD37YPFYKXY","created_at":"2026-07-14T00:18:35.697502+00:00"},{"alias_kind":"pith_short_8","alias_value":"K4QC7AD3","created_at":"2026-07-14T00:18:35.697502+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/K4QC7AD37YPFYKXY4NXGJM5SJA","json":"https://pith.science/pith/K4QC7AD37YPFYKXY4NXGJM5SJA.json","graph_json":"https://pith.science/api/pith-number/K4QC7AD37YPFYKXY4NXGJM5SJA/graph.json","events_json":"https://pith.science/api/pith-number/K4QC7AD37YPFYKXY4NXGJM5SJA/events.json","paper":"https://pith.science/paper/K4QC7AD3"},"agent_actions":{"view_html":"https://pith.science/pith/K4QC7AD37YPFYKXY4NXGJM5SJA","download_json":"https://pith.science/pith/K4QC7AD37YPFYKXY4NXGJM5SJA.json","view_paper":"https://pith.science/paper/K4QC7AD3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.09786&json=true","fetch_graph":"https://pith.science/api/pith-number/K4QC7AD37YPFYKXY4NXGJM5SJA/graph.json","fetch_events":"https://pith.science/api/pith-number/K4QC7AD37YPFYKXY4NXGJM5SJA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/K4QC7AD37YPFYKXY4NXGJM5SJA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/K4QC7AD37YPFYKXY4NXGJM5SJA/action/storage_attestation","attest_author":"https://pith.science/pith/K4QC7AD37YPFYKXY4NXGJM5SJA/action/author_attestation","sign_citation":"https://pith.science/pith/K4QC7AD37YPFYKXY4NXGJM5SJA/action/citation_signature","submit_replication":"https://pith.science/pith/K4QC7AD37YPFYKXY4NXGJM5SJA/action/replication_record"}},"created_at":"2026-07-14T00:18:35.697502+00:00","updated_at":"2026-07-14T00:18:35.697502+00:00"}