{"id":"W7082246090","doi":"10.48448/9epa-4248","title":"Towards Safety Reasoning in LLMs: AI-agentic Deliberation for Policy-embedded CoT Data Creation","year":2025,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Deliberation; Process (computing); Recipe; Robustness (evolution); Iterative and incremental development; Process safety","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003925513,0.001373549,0.0004520297,0.0009246308,0.0006959159,0.002042299,0.002446699,0.001525486,0.006464215],"category_scores_gemma":[0.0229739,0.0005312726,0.001259145,0.0004034031,0.001877487,0.003133032,0.004672603,0.003351572,0.0020049],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001011443,"about_ca_system_score_gemma":0.001917867,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003012335,"about_ca_topic_score_gemma":0.006678205,"domain_scores_codex":[0.9971733,0.001305803,0.0001918175,0.0006344941,0.0005313243,0.0001631136],"domain_scores_gemma":[0.9912362,0.004671913,0.0004949964,0.002524626,0.0007921373,0.0002800896],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0009014027,0.0006882721,0.009595691,0.001279929,0.0002756637,0.0006904107,0.003090717,0.3391898,0.02879971,0.06894149,0.02590098,0.520646],"study_design_scores_gemma":[0.00007369636,0.0001095112,0.000590903,0.0001236625,0.00003460813,0.000120493,0.000338272,0.8870054,0.01675544,0.07721637,0.0175892,0.0000423507],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01776638,0.0001835781,0.9616773,0.0007461872,0.0001054607,0.0002757055,0.0006675839,0.01459308,0.003984825],"genre_scores_gemma":[0.2700036,0.0001175223,0.723558,0.0006711549,0.00003885745,0.0003412349,0.001861742,0.0009845259,0.002423398],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.006464215,"threshold_uncertainty_score":0.02162492,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0284676570974151,"score_gpt":0.3334963186949292,"score_spread":0.3050286615975141,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}