{"id":"W4415947615","doi":"10.1007/978-3-032-10489-2_36","title":"Evaluating AI Agents for Cyber Defense: A Comparison of Deep Reinforcement Learning and LLM Approaches","year":2025,"lang":"en","type":"book-chapter","venue":"Lecture notes in computer science","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Dalhousie University","funders":"","keywords":"Reinforcement learning; Adaptability; Curiosity; Robustness (evolution); Transparency (behavior); Autonomous agent; Intelligent agent; Deep learning","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":[0.001929818,0.0005339555,0.0008393158,0.0007732896,0.0004805462,0.0003285546,0.001873758,0.000315988,0.00001109562],"category_scores_gemma":[0.0008099956,0.0005215614,0.0001514709,0.0005004148,0.0005317194,0.0004237067,0.00221785,0.001137458,0.00000200867],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000244223,"about_ca_system_score_gemma":0.000397788,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001754058,"about_ca_topic_score_gemma":0.00001988688,"domain_scores_codex":[0.9959021,0.00009736206,0.0008667896,0.001458018,0.001050759,0.0006249677],"domain_scores_gemma":[0.996696,0.001396188,0.0007076573,0.0007623549,0.0003212126,0.0001165996],"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":[0.00001077444,0.00001381547,0.0003407855,0.0001252386,0.00001841382,0.000002155771,0.001311198,0.6001282,0.0000150272,0.01755566,0.000004463857,0.3804743],"study_design_scores_gemma":[0.0005203321,0.000380053,0.0000409833,0.0004930186,0.00003283083,0.000007677092,0.000002067939,0.976999,0.0002847314,0.02029204,0.0004927546,0.000454477],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00008843315,0.000541965,0.9949688,0.0003462229,0.0007779627,0.0008096636,8.413509e-7,0.00009490993,0.002371219],"genre_scores_gemma":[0.3686289,0.000009782729,0.6302453,0.0003100724,0.0001559372,0.00002590816,0.000009014135,0.00002800144,0.0005870271],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3800198,"threshold_uncertainty_score":0.9997236,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06335727882812493,"score_gpt":0.3479322205925678,"score_spread":0.2845749417644429,"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."}}