AI chatbots like ChatGPT and Claude found unsafe for medical advice in US

A recent survey shows that roughly one‑third of Americans turn to artificial‑intelligence chatbots for health information. Experts warn that in more than half of emergency‑type queries the bots fail to provide correct guidance.

Researchers evaluating OpenAI’s ChatGPT Health and Anthropic’s Claude for Healthcare found that the systems often give confident but inaccurate answers. In over 50 % of cases involving potential respiratory failure or severe diabetic complications, ChatGPT did not advise users to seek emergency care, instead suggesting they monitor symptoms at home. Similar shortcomings were noted in other independent studies, prompting health‑policy groups to advise caution.

Both companies stress that their products are not intended to replace medical professionals. OpenAI says its models are continuously refined with physician feedback, while Anthropic highlights privacy safeguards and integration with medical records. Nonetheless, the combination of high usage, costly or inaccessible traditional care, and the bots’ persuasive tone raises concerns about patient safety.