{"id":"W4414590635","doi":"10.48550/arxiv.2509.05367","title":"Between a Rock and a Hard Place: The Tension Between Ethical Reasoning and Safety Alignment in LLMs","year":2025,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Law, Economics, and Judicial Systems","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Tsinghua University; Strong; Institute for Catastrophic Loss Reduction","keywords":"Vulnerability (computing); Frame (networking); Ethical theories; Motivated reasoning; Poison control; Ethical issues; Action (physics)","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0102757,0.0005927184,0.0008750373,0.0009459975,0.001578336,0.006516211,0.002003135,0.002090908,0.005494772],"category_scores_gemma":[0.04431304,0.0007870503,0.001429933,0.0005834767,0.009111217,0.009758003,0.00803946,0.005435418,0.0009284702],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003156176,"about_ca_system_score_gemma":0.002581264,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003547068,"about_ca_topic_score_gemma":0.003435801,"domain_scores_codex":[0.9879027,0.008311405,0.0004656972,0.001097425,0.001438151,0.0007846594],"domain_scores_gemma":[0.9703569,0.01721468,0.002221987,0.008030184,0.001273552,0.0009027196],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001769026,0.00006720097,0.00265201,0.00008764433,0.00006356286,0.0002368906,0.002605667,0.09134992,0.002084932,0.8739362,0.001736688,0.02500236],"study_design_scores_gemma":[0.00002802906,0.00003483562,0.0003449428,0.0000395984,0.00003187518,0.00005738075,0.0004701041,0.2531238,0.002707779,0.7396503,0.003479668,0.00003159728],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1181492,0.0001356575,0.8587118,0.005225674,0.00004997603,0.00009386767,0.0001476006,0.001669388,0.01581684],"genre_scores_gemma":[0.9166622,0.00004846602,0.08017562,0.0003777654,0.0000350607,0.00008504785,0.00008911663,0.0003348607,0.002191807],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0102757,"threshold_uncertainty_score":0.05434376,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07003019765670702,"score_gpt":0.1925403365103339,"score_spread":0.1225101388536269,"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."}}