Stanford, OpenAI Build Watchdog for Healthcare AI Agents
Stanford Medicine researchers are getting up to $14.9 million from the Advanced Research Projects Agency for Health to build a system that polices autonomous AI agents in cardiovascular care, according to Becker's Hospital Review. The project is called STEWARD, short for safety, trust, evaluation and watchdog for embedded agentic care delivery. OpenAI is a partner on the effort.
STEWARD is a three-stage supervisory system meant to monitor AI agents as they operate, catching errors and unsafe behavior before it reaches patients. The focus on agentic AI is notable: these are systems that act on their own rather than simply answering a prompt, which raises the stakes for oversight in clinical settings.
The funding signals a shift in how the industry is approaching AI safety. Rather than trusting individual models, health systems and regulators are investing in independent monitoring layers. As hospitals begin deploying agents that can take clinical actions, watchdog infrastructure like STEWARD could become a prerequisite for safe, scaled adoption.
Sources