Descriptions:
Karan Singhal, who leads Health AI at OpenAI and was recently named to the Time 100 Health list, joins the Cognitive Revolution to discuss the company’s medical AI strategy and two products launching in early 2026: ChatGPT Health, which connects to electronic medical record systems and consumer wearables, and a physician-facing ChatGPT product for clinical settings.
Singhal provides detailed context for Healthbench, OpenAI’s internal evaluation suite containing 49,000 individual criteria spanning medical accuracy, care escalation judgment, demographic and epidemiological adaptability, and model calibration. GPT-4o scored 0% on the Healthbench Hard subset when the benchmark was first created; current models have reached 40%. He also describes a first-of-its-kind randomized controlled trial conducted with Kenya’s Penda Health system, which showed a statistically significant improvement in patient outcomes when physicians used AI co-pilots — an early signal that AI assistance can move clinical metrics, not just accuracy benchmarks. Scale context: 230 million people currently use ChatGPT for health questions on a weekly basis.
The conversation covers how OpenAI trains models to acknowledge uncertainty, browse for current clinical information, and calibrate responses to local epidemiological conditions. Singhal describes OpenAI’s stated plan to make ChatGPT Health globally available for free with no advertising, framing it as a form of universal basic intelligence. The episode also addresses privacy and security architecture for health data, the relationship between health AI and broader alignment work, and Singhal’s view that 2026 will mark the year AI becomes a standard part of medical practice.
📺 Source: Cognitive Revolution “How AI Changes Everything” · Published February 25, 2026
🏷️ Format: Interview
