{"id":"W4407012542","doi":"10.48550/arxiv.2501.18536","title":"Illusions of Relevance: Arbitrary Content Injection Attacks Deceive Retrievers, Rerankers, and LLM Judges","year":2025,"lang":"en","type":"preprint","venue":"ArXiv.org","topic":"Digital and Cyber Forensics","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Relevance (law); Illusion; Content (measure theory); Psychology; Social psychology; Cognitive psychology; Political science; Law; Mathematics","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0001914557,0.0003083636,0.0004577709,0.0001635117,0.0001280912,0.0000984131,0.0007404237,0.000295967,0.000005506313],"category_scores_gemma":[0.0002050351,0.0002872472,0.000175304,0.0004036992,0.0002341569,0.000407352,0.002084959,0.0006156488,0.00001165958],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008465517,"about_ca_system_score_gemma":0.0002601235,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001611402,"about_ca_topic_score_gemma":0.00003794775,"domain_scores_codex":[0.9980888,0.00005733792,0.0004747915,0.0007808096,0.0003195595,0.0002786832],"domain_scores_gemma":[0.9982061,0.0001813401,0.0002969585,0.0009159439,0.0002789121,0.0001207077],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002922186,0.000896547,0.7460992,0.001368344,0.0009731404,0.0001640476,0.00679869,0.0005002091,0.001290533,0.08132809,0.03681893,0.1234701],"study_design_scores_gemma":[0.002358942,0.001105222,0.8548048,0.003168605,0.000289215,0.00009529807,0.0008643066,0.003396913,0.01640441,0.0810751,0.0341538,0.002283349],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9740878,0.001575695,0.008157961,0.001628506,0.001997629,0.0004153998,0.0001003586,0.0002099155,0.01182677],"genre_scores_gemma":[0.9910245,0.001226692,0.003375127,0.0008671631,0.0000989012,0.00002146594,0.00005737781,0.00001528522,0.003313517],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1211867,"threshold_uncertainty_score":0.999958,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04721395062880226,"score_gpt":0.256913845652999,"score_spread":0.2096998950241968,"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."}}