{"id":"W7117255613","doi":"10.20944/preprints202512.2112.v1","title":"Calibrated Trust in AI for Security Operations: A Conceptual Framework for Analyst–AI Collaboration","year":2025,"lang":"","type":"preprint","venue":"Preprints.org","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Tetra Tech (Canada); Ontario Tech University","funders":"","keywords":"Trustworthiness; Conceptual framework; Key (lock); Foundation (evidence); Automation; Empirical research; Conceptual model; Computational trust","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":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.003952381,0.001184481,0.001810485,0.0009388105,0.001073704,0.0007848846,0.003503946,0.001912728,0.0004331905],"category_scores_gemma":[0.01260938,0.00145319,0.0006193562,0.003280865,0.0005247989,0.001818554,0.004481113,0.003618964,0.00009109115],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00112461,"about_ca_system_score_gemma":0.00488297,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008292682,"about_ca_topic_score_gemma":0.0008217907,"domain_scores_codex":[0.9896341,0.001509543,0.002674222,0.004095477,0.0008230832,0.001263583],"domain_scores_gemma":[0.9908872,0.001994347,0.0009574645,0.003244412,0.002578538,0.0003380977],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0004584333,0.0003675155,0.07995643,0.0005150005,0.0002932905,0.000006725101,0.01485252,0.6407044,0.0001821389,0.2619591,0.00009016203,0.0006143192],"study_design_scores_gemma":[0.002487077,0.00011174,0.006224952,0.00123902,0.0002181842,0.000002344032,0.001280215,0.9085077,0.003471588,0.07136561,0.003858193,0.001233346],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08813784,0.0002060508,0.8899077,0.008990505,0.003719919,0.007987373,0.0004158188,0.0003284525,0.000306388],"genre_scores_gemma":[0.884944,0.00008761222,0.1068121,0.002105394,0.0006971738,0.00394145,0.0004001468,0.00007658052,0.0009356306],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7968061,"threshold_uncertainty_score":0.999383,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07830353351326334,"score_gpt":0.4023939926707834,"score_spread":0.32409045915752,"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."}}