{"id":"W7082279302","doi":"10.48448/sg6p-3t40","title":"Exploiting Instruction-Following Retrievers for Malicious Information Retrieval","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":"McGill University; Mila - Quebec Artificial Intelligence Institute","funders":"","keywords":"Work (physics); Information system; Information protection policy; Risk assessment","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":[0.002867866,0.001242077,0.000826625,0.0014604,0.0006285653,0.001951711,0.001078349,0.001367241,0.002911171],"category_scores_gemma":[0.02844805,0.000339887,0.0006083429,0.0008625079,0.00113812,0.003849237,0.002574836,0.001218914,0.003258287],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005232545,"about_ca_system_score_gemma":0.001053702,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003148552,"about_ca_topic_score_gemma":0.005139689,"domain_scores_codex":[0.9966067,0.001175625,0.000323801,0.0004436786,0.001195609,0.0002545558],"domain_scores_gemma":[0.9841148,0.007904492,0.001341938,0.004863499,0.001436851,0.000338406],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00320214,0.000914153,0.07902256,0.003505874,0.0005910241,0.002673217,0.005642063,0.03552253,0.1568401,0.007895349,0.04123633,0.6629547],"study_design_scores_gemma":[0.0003224048,0.004160285,0.03281681,0.0003884652,0.0006371827,0.00597235,0.003653313,0.6107964,0.2388378,0.01734007,0.08462774,0.0004472217],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.808395,0.005537132,0.1170467,0.001495546,0.0003333371,0.0006612421,0.002476146,0.04831759,0.01573735],"genre_scores_gemma":[0.9331565,0.0006296394,0.05594216,0.0006392414,0.0001006915,0.0001364332,0.00266638,0.001367169,0.005361782],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003148552,"threshold_uncertainty_score":0.01516688,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0122172574850315,"score_gpt":0.2508977393615315,"score_spread":0.2386804818765,"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."}}