{"id":"W4415230414","doi":"10.1609/aies.v8i1.36596","title":"Incident Analysis for AI Agents","year":2025,"lang":"en","type":"article","venue":"Proceedings of the AAAI/ACM Conference on AI Ethics and Society","topic":"Anomaly Detection Techniques and Applications","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institute on Governance","funders":"","keywords":"Documentation; Incident report; Incident management; Private information retrieval; Information system; Incident response","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":[],"consensus_categories":[],"category_scores_codex":[0.0007660441,0.0001069881,0.0001802818,0.00005869419,0.0005065417,0.0002010538,0.00104601,0.0001705559,0.000004040416],"category_scores_gemma":[0.0002083781,0.00008085834,0.0002691967,0.0007617976,0.0001327757,0.0001390996,0.0005214323,0.0004960753,5.589336e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003083786,"about_ca_system_score_gemma":0.0001075656,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000334598,"about_ca_topic_score_gemma":0.000003367579,"domain_scores_codex":[0.999111,0.000008470617,0.0002103543,0.0003132726,0.000215644,0.0001412444],"domain_scores_gemma":[0.998686,0.0001305186,0.0001601033,0.0002867225,0.0006977342,0.00003888472],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000002741343,0.00003489272,0.002496865,0.00006395107,0.0001260995,6.965025e-9,0.001474072,0.000003674557,0.001280576,0.9864706,0.005925027,0.0021215],"study_design_scores_gemma":[0.00050387,0.0001782141,0.03359194,0.0001894879,0.0003660383,0.000001041882,0.00106339,0.2005195,0.04157161,0.7024345,0.0191769,0.0004034406],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.06638153,0.00004994402,0.7830817,0.1436772,0.0001431452,0.0008811284,0.00001863435,0.0002054692,0.005561247],"genre_scores_gemma":[0.9760599,0.0002570701,0.01578766,0.006782554,0.0000124376,0.00009316803,6.423733e-7,0.000003517366,0.001003061],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9096783,"threshold_uncertainty_score":0.389596,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06055215157506705,"score_gpt":0.3572805285963317,"score_spread":0.2967283770212647,"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."}}