{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001525542,0.0002634441,0.0003033614,0.0006972694,0.0002617663,0.000365095,0.003135452,0.0002280377,0.0001010876],"category_scores_gemma":[0.001901305,0.0002570284,0.00003856298,0.001686588,0.0002934349,0.0005958494,0.001168514,0.0002339688,0.00001543609],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002319531,"about_ca_system_score_gemma":0.002743094,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007853529,"about_ca_topic_score_gemma":0.0005343291,"domain_scores_codex":[0.997345,0.00005614116,0.0004203248,0.001188182,0.0004684311,0.0005218903],"domain_scores_gemma":[0.9977154,0.0000821707,0.0002542318,0.001593997,0.0002559468,0.00009830941],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0000297126,0.000255473,0.000163728,0.0005150248,0.00005496495,0.0000174489,0.0006921543,0.004163259,0.001717881,0.5252112,0.1913292,0.2758499],"study_design_scores_gemma":[0.0004927873,0.00003633951,0.0001396213,0.000436085,0.0000158053,0.000007307137,0.00004795354,0.699218,0.0005373865,0.01527324,0.2834111,0.0003843038],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"other","genre_scores_codex":[0.000003190313,0.0001063406,0.6828234,0.00679339,0.000375928,0.0004460429,0.0000651911,0.0001858907,0.3092006],"genre_scores_gemma":[0.0290816,0.0001706009,0.358472,0.002609969,0.0009381038,0.00008630776,0.001483585,0.00005578328,0.6071021],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6950548,"threshold_uncertainty_score":0.9999882,"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."}}