Stanford and Grow Therapy Partner to Benchmark AI Safety in Psychiatry

The collaboration moves beyond hypothetical testing by utilizing de-identified patient scenarios captured from Grow’s existing AI-coach platform. A panel of psychiatrists will evaluate how leading AI models navigate these sensitive interactions, specifically testing whether specialized safety reviewers effectively mitigate potential harm. The goal is to provide a standardized framework for the industry, helping to define what constitutes acceptable AI behavior when clinical outcomes are at stake.

Grow Therapy, which operates a network of 26,000 mental health professionals, brings its own multi-layered safety infrastructure to the project. While the company emphasizes the necessity of keeping licensed clinicians in the loop, the research addresses the reality that many individuals rely on general-purpose LLMs for counseling. By integrating real-world clinical experience with academic rigor, the researchers intend to create a shared yardstick for safety that could reduce risks for users across all AI platforms.