analiza aiqbench
This research introduces NOHARM, a rigorous benchmark evaluating clinical safety risks in large language models used for medical consultations. The study reveals that severe harm potential exists in over 20% of cases across 28 tested models, with errors of omission being the primary failure mode, while also noting a gap between AI assistance and actual physician adoption rates. These findings underscore the critical need to decouple clinical safety metrics from general intelligence scores before deploying LLMs in high-stakes healthcare environments.
First, do NOHARM: towards clinically safe large language models
arXiv:2512.01241v3 Announce Type: replace-cross Abstract: Large language models (LLMs) are routinely used by physicians and patients for medical advice, yet their clinical safety profiles remain poorly characterized. We present NOHARM (Numerous Options Harm Assessment for Risk in Medicine), a 1,100-task benchmark of primary care-to-specialist consultation cases to measure the frequency and severity of harm from LLM-generated medical recommendations. NOHARM covers 10 specialties, with 12,747 expert annotations for 4,249 clinical management options. Across 28 LLMs, recommendations carried the potential for severe harm in up to 22.6% of cases, with errors of omission accounting for more than 80% of severe errors. In a randomized trial of 101 generalist physicians, human benchmark performance significantly improved with AI assistance, yet physicians remained far from realizing the potential of AI tools, frequently ignoring essential advice surfaced by AI. Safety performance tracked general-intelligence and medical-knowledge benchmarks across the full range of models but decoupled at the frontier. Despite strong performance on existing evaluations, widely used AI models can produce medical advice with the potential for severe harm at non-trivial rates, highlighting the importance of explicit measurement of clinical safety.
Authors:David Wu, Fateme Nateghi Haredasht, Saloni Kumar Maharaj, Priyank Jain, Jessica Tran, Matthew Gwiazdon, Arjun Rustagi, Jenelle Jindal, Jacob M. Koshy, Vinay Kadiyala, Anup Agarwal, Bassman Tappuni, Brianna French, Sirus Jesudasen, Christopher V. Cosgriff, Rebanta Chakraborty, Jillian Caldwell, Susan Ziolkowski, David J. Iberri, Robert Diep, Rahul S. Dalal, Kira L. Newman, Kristin Galetta, J. Carl Pallais, Nancy Wei, Kathleen M. Buchheit, David I. Hong, Vartan Pahalyants, Ernest Y. Lee, Allen Shih, Tamara B. Kaplan, Vishnu Ravi, Sarita Khemani, Thomas A. Buckley, April S. Liang, Daniel Shirvani, Advait Patil, Nicholas Marshall, Kanav Chopra, Joel Koh, Adi Badhwar, Anastasia Perez, Austin J. Schoeffler, Mahbuba Tusty, Chase M. Walton, Liam G. McCoy, David J. H. Wu, Yingjie Weng, Sumant Ranji, Kevin Schulman, Nigam H. Shah, Jason Hom, Arnold Milstein, Arjun K. Manrai, Adam Rodman, Jonathan H. Chen, Ethan Goh
Abstract:Large language models (LLMs) are routinely used by physicians and patients for medical advice, yet their clinical safety profiles remain poorly characterized. We present NOHARM (Numerous Options Harm Assessment for Risk in Medicine), a 1,100-task benchmark of primary care-to-specialist consultation cases to measure the frequency and severity of harm from LLM-generated medical recommendations. NOHARM covers 10 specialties, with 12,747 expert annotations for 4,249 clinical management options. Across 28 LLMs, recommendations carried the potential for severe harm in up to 22.6% of cases, with errors of omission accounting for more than 80% of severe errors. In a randomized trial of 101 generalist physicians, human benchmark performance significantly improved with AI assistance, yet physicians remained far from realizing the potential of AI tools, frequently ignoring essential advice surfaced by AI. Safety performance tracked general-intelligence and medical-knowledge benchmarks across the full range of models but decoupled at the frontier. Despite strong performance on existing evaluations, widely used AI models can produce medical advice with the potential for severe harm at non-trivial rates, highlighting the importance of explicit measurement of clinical safety.
Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
arXiv:2512.01241 [cs.CY]
(or arXiv:2512.01241v3 [cs.CY] for this version)
Text integral sub licența deschisă a publicației · 329 cuvinte
Subiecte conexe
ChatHealthAI: Aligning Electronic Health Record Representations with Large Language Models for Grounded Clinical Reasoning
arXiv · 9 iun. 2026
ChatHealthAI: Aligning Electronic Health Record Representations with Large Language Models for Grounded Clinical Reasoning
arXiv · 3 iun. 2026
Safety That Does Not Transfer: Cross-Lingual Clinical Correctness Drift in Deployable Medical Language Models
arXiv · 21 iul. 2026
Same Facts, Different Diagnosis: Measuring and Mitigating Narrative Anchoring in Clinical Language Models
arXiv · 31 iul. 2026