{"paper":{"title":"Monitorability Is Trainable: Post-Training for Chain-of-Thought Monitorability","arxiv_id":"2610.29734","primary_cat":"cs.LG","submitted_at":"2026-10-21T13:35:57+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Chain-of-thought (CoT) monitoring is one of the few practical tools available for overseeing capable AI systems, but it is fragile: a model's reasoning is only useful to a monitor if it remains coupled to what the model actually does.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.07,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.12,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.99,"id":"s1","own":false,"role":"background","text":"This coupling can be degraded during training through, for example, CoT compression, optimization against a monitor, and out-of-context generalization from synthetic documents.","hedge":{"p":0.53,"label":"hedged_possibility"},"own_p":0.15,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.01,"method":0.0,"result":0.0,"background":0.99,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.63,"id":"s2","own":true,"role":"method","text":"In this work, we propose an unsupervised post-training objective based on maximizing mutual information between CoT and the output for improving CoT monitorability.","hedge":{"p":0.9,"label":"asserted"},"own_p":0.85,"main_claim_p":0.09,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.63,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.37,"interpretation":0.0}},{"i":3,"p":1.0,"id":"s3","own":true,"role":"result","text":"Across three model organisms with degraded CoT, it recovers most or all of the lost monitorability.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.12,"quantitative":true,"quantitative_p":0.68,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":4,"p":1.0,"id":"s4","own":true,"role":"result","text":"Applied to undegraded policies, it further improves judged monitorability, improves prediction of unsafe behavior by CoT monitors and activation probes, and increases the prevalence of consequential reasoning.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.11,"quantitative":false,"quantitative_p":0.3,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":5,"p":1.0,"id":"s5","own":true,"role":"interpretation","text":"These results suggest that monitorability should be treated as a trainable attribute of the policy.","hedge":{"p":1.0,"label":"hedged_possibility"},"own_p":0.58,"main_claim_p":0.88,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":1.0}},{"i":6,"p":0.59,"id":"s6","own":false,"role":"interpretation","text":"A dedicated monitorability post-training stage can extend current practices that protect CoT from direct optimization pressure.","hedge":{"p":0.79,"label":"hedged_possibility"},"own_p":0.23,"main_claim_p":0.1,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.39,"contribution":0.02,"interpretation":0.59}}]},"compiled":{"main_claim":{"p":1.0,"text":"These results suggest that monitorability should be treated as a trainable attribute of the policy.","location":"abstract s5","hedge_strength":{"p":1.0,"label":"hedged possibility"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":5611,"cost_usd_reported":0.0},"compiled_at":"2026-10-22","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.07,"states_limitation":0.18,"human_participants":0.04,"controlled_comparison":0.57,"randomized_assignment":0.02,"states_quantitative_result":0.09,"compares_with_prior_results":0.46,"mentions_code_or_data_availability":0.03},"evidential_basis":{"p":0.94,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.94,"simulation_or_computation":0.06}},"contribution_type":{"p":0.99,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":0.99,"empirical_study":0.01,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":1.0,"text":"In this work, we propose an unsupervised post-training objective based on maximizing mutual information between CoT and the output for improving CoT monitorability.","location":"abstract s2"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":7,"sentences_in_abstract":7},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}