OpenAI’s new ChatGPT Health feature lets users link Apple Health data to ask medical questions. A Washington Post reporter fed the bot ten years of his Apple Watch data—29 million steps and 6 million heartbeats—and asked it to grade his cardiac health. The AI returned an F today.

Doctors and cardiologists slammed the result. Fowler’s own physician said he was at low risk and that insurance would likely not cover extra tests. Scripps Research’s Eric Topol called the analysis baseless, warning that consumers should ignore AI medical advice until it proves reliable for today in the future.

OpenAI and Anthropic’s Claude also struggled. Claude graded the same data a C but failed to account for Apple Watch limitations. Both firms claim their tools are advisory, not diagnostic, yet the FDA’s recent stance—allowing AI to make medical claims without review—raises regulatory concerns.

OpenAI says it will refine consistency before a wider rollout. The beta launch, limited to a waitlist, offers a chance to learn from real‑world data. Experts warn that inconsistent grading could mislead patients, underscoring the need for clear FDA oversight and transparent AI health claims for users today.