{"id":"W4415300695","doi":"10.1038/s41746-025-02008-z","title":"When helpfulness backfires: LLMs and the risk of false medical information due to sycophantic behavior","year":2025,"lang":"en","type":"article","venue":"npj Digital Medicine","topic":"Topic Modeling","field":"Computer Science","cited_by":52,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Google Research; National Center for Advancing Translational Sciences; European Commission; Harvard Catalyst; National Cancer Institute; National Institutes of Health; Harvard University; Patient-Centered Outcomes Research Institute","keywords":"Helpfulness; Vulnerability (computing); Consistency (knowledge bases); Suspect; Risk assessment; Benchmark (surveying); Baseline (sea); Sophistication","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":[0.01889143,0.00116463,0.0009437573,0.0009743046,0.0006881268,0.002060887,0.0013345,0.002308388,0.001460534],"category_scores_gemma":[0.1003525,0.0005683345,0.0007162965,0.0005043191,0.001566208,0.003305209,0.002129895,0.00378956,0.0007419727],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001092063,"about_ca_system_score_gemma":0.002000579,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004315872,"about_ca_topic_score_gemma":0.004589023,"domain_scores_codex":[0.990975,0.005861513,0.0005994758,0.001354293,0.0008890812,0.0003206324],"domain_scores_gemma":[0.8675652,0.1143137,0.005118197,0.008489057,0.003163172,0.001350676],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006520025,0.001345482,0.2990824,0.001619555,0.0008081608,0.002735994,0.009282188,0.241708,0.03010434,0.00940052,0.02022589,0.3771674],"study_design_scores_gemma":[0.0003298731,0.001579497,0.03272663,0.0003836721,0.0003453647,0.001782325,0.001800678,0.9005942,0.02277718,0.02881924,0.008665741,0.0001956746],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8783748,0.001517654,0.1038813,0.005501994,0.0002111625,0.0002421817,0.0008713725,0.006339303,0.003060108],"genre_scores_gemma":[0.9753379,0.0001732988,0.02178027,0.0009421756,0.0000607288,0.00005914526,0.0007111019,0.0001759933,0.0007594891],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01889143,"threshold_uncertainty_score":0.09990865,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.008338394089495093,"score_gpt":0.2471250392254192,"score_spread":0.2387866451359241,"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."}}