Doximity Outranks OpenEvidence, Frontier Models in Independent Stanford-Harvard Study of Clinical AI Safety - Financial Times

Doximity Ask, a clinical AI tool, ranked first in the NOHARM benchmark for clinical AI safety, according to a study by ARISE from Stanford and Harvard. The study evaluated AI models on real-world clinical samples, and Doximity Ask outperformed leading frontier AI models.

The NOHARM benchmark was designed to assess the safety of AI in clinical settings, focusing on real-world scenarios. Doximity Ask, which is built on a large language model, achieved the highest safety scores among the models tested.

The ARISE study is described as an independent evaluation of clinical AI safety. The results highlight that domain-specific models may perform better in safety-critical tasks than general-purpose frontier models. Doximity, the company behind the tool, markets it as a clinical workflow product used by healthcare professionals for tasks like prior authorization and clinical knowledge retrieval.