{"id":"W7128225249","doi":"10.1162/tacl.a.27","title":"Investigating Adversarial Trigger Transfer in Large Language Models","year":2025,"lang":"en","type":"article","venue":"Transactions of the Association for Computational Linguistics","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Institute for Advanced Research","funders":"","keywords":"Adversarial system; Security token; Offensive; Robustness (evolution); Language model; Threat model; Language understanding","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":[0.0007039987,0.0001064997,0.000187854,0.0001801754,0.0002465087,0.00004071211,0.0004957511,0.0001020365,0.000002861647],"category_scores_gemma":[0.001693452,0.0001032639,0.0001577944,0.0006825603,0.00002433973,0.0001111644,0.00002500827,0.0002571636,8.334024e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002576469,"about_ca_system_score_gemma":0.0002638574,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005206128,"about_ca_topic_score_gemma":0.00004830924,"domain_scores_codex":[0.9987255,0.0001227523,0.000428612,0.0001992527,0.000328971,0.0001949527],"domain_scores_gemma":[0.9980693,0.001138681,0.0001508588,0.0001852582,0.0004330667,0.00002282755],"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":[0.000005678086,0.00004992519,0.0007924619,0.00002173623,0.00003476313,8.61038e-8,0.0008322089,0.7214884,0.00001182322,0.2764742,0.0000423486,0.0002464317],"study_design_scores_gemma":[0.001299258,0.00001229619,0.001663238,0.00004733353,0.00003909627,1.446556e-7,0.00007316389,0.9168312,0.0001426082,0.07921344,0.0005862161,0.00009200824],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.001847487,0.00002163003,0.9939311,0.0007291547,0.001305853,0.0003017184,0.00005992154,0.00005891227,0.001744274],"genre_scores_gemma":[0.9368368,9.388809e-7,0.06225672,0.0001529149,0.00007603274,0.00001881908,0.00001389511,0.000007976543,0.0006358722],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9349893,"threshold_uncertainty_score":0.421098,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01217664085012492,"score_gpt":0.2804505190115679,"score_spread":0.2682738781614429,"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."}}