First, do NOHARM: a medical safety benchmark and randomized study of physician and AI teaming on clinical consultations
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
Why It Matters
What makes this one worth your time
Understanding and mitigating the risks of AI in clinical settings is crucial for ensuring patient safety and improving healthcare outcomes.
NOHARM benchmark reveals safety risks in AI medical consultations and highlights the potential of human-AI collaboration.
Summary
The paper introduces NOHARM, a benchmark designed to evaluate the safety of large language models (LLMs) and clinical AI tools in medical consultations, revealing significant potential for harm in AI-generated recommendations. It also presents a study showing that AI assistance can improve physician performance, though human-AI teaming has untapped potential.
Key contributions
- Development of the NOHARM benchmark for assessing clinical safety of AI tools.
- Randomized study demonstrating the impact of AI assistance on physician performance.
- Public availability of the benchmark and leaderboard for ongoing AI evaluation.
Notable insights
- Errors of omission are a major source of severe harm in AI-generated medical advice.
- Human-AI teaming shows potential for superior performance but requires better integration of AI recommendations.
Possible limitations
- Not stated in the abstract
Abstract
arXiv:2512.01241v4 Announce Type: replace-cross Abstract: Large language models (LLMs) and medical AI tools 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 potentially harmful errors from LLM-generated medical consultation recommendations. NOHARM covers 10 specialties, with 12,747 expert annotations for 4,249 clinical management options. Across 20 notable LLMs and 4 widely used retrieval-augmented generation (RAG) clinical AI tools, direct application of recommendations carried potential for severe harm in up to 24.6% of cases, with errors of omission accounting for more than 80% of severe errors. Harm potential was not uniform across systems, with clinical AI tools outperforming generalist LLMs, and multi-agent AI teaming further improving performance in generalist models. In a randomized study of 101 U.S.-licensed generalist physicians, AI assistance improved physician performance compared to conventional resources. However, AI-assisted physicians frequently omitted valuable AI-generated recommendations and still scored lower than many AI systems alone. Had those recommendations been incorporated, combined human-AI responses would have outperformed both the human and AI system as used, suggesting complementary strengths and unrealized potential in human-AI teaming. Collectively, these results show that despite strong performance on medical knowledge benchmarks, widely used AI tools can produce medical consultation advice with the potential for severe harm, and highlight the need for explicit measurement of clinical safety. The benchmark and leaderboard are publicly available to support ongoing evaluation and improvement of AI systems used for clinical care.