The piece argues that releasing ChatGPT Health without thorough safety validation is like shipping a vehicle that isn’t road-ready, underscoring the need for accountability and robust safeguards before public health tools go live.

Ultimately, public health could become the final test of ChatGPT Health’s usefulness and safety unless accountability and rigorous safeguards improve prior to broad deployment.

There is a strong warning against launching health AI products before proper validation, drawing a parallel to unsafe consumer products that reach the market without adequate safety testing.

ChatGPT Health aims to access medical records and lab results to explain visits, interpret results, and aid doctor–patient discussions, but this clashes with public expectations for definitive medical guidance.

The piece traces the shift from purely parametric LLMs to retrieval-augmented generation (RAG), which enables real-time data access but raises concerns about quality assessment and reliance on AI for clinical decisions.

Tests show ChatGPT Health’s triage performance is worrying, with a high rate of under-triage in true emergencies and only about half the time alignment with professional decisions.

Structured evaluations indicate roughly 50% agreement with standard emergency triage decisions, highlighting significant risk if used for first-contact triage.

OpenAI uses Health Bench—an assessment framework developed with hundreds of physicians—to audit and improve ChatGPT Health, though data from these assessments have not yet been published and progress remains in progress.

OpenAI quietly rolled out ChatGPT Health to a select group of US adults, linking to personal medical records and Apple wearable data to tailor conversations, while clarifying it is not a diagnostic tool or substitute for professional advice.

OpenAI emphasizes ChatGPT Health is not for diagnosis or treatment and should not replace medical judgment, yet public use trends risk treating it as a medical decision aid.

Multiple studies show health AI chatbots can give problematic or unsafe responses, with mis-triage in emergencies and inconsistent medical guidance, signaling limited safety and effectiveness evidence.

There is broad industry concern about health AI safety and calls for stronger safeguards, independent oversight, and even slowing advanced AI development to ensure safety.