Can You Trust Your AI Doctor? Don’t Bet Your Health on It

Subscribe to our Telegram channel for the latest stories and updates.

AI chatbots can be made to lie about health, and they do it with unsettling ease. Australian researchers tested major language models and found that most deliver false medical information on command. The output often includes fake citations, scientific terms, and a tone that mimics clinical authority.

GPT-4o, Gemini 1.5 Pro, Llama 3.2, Grok, and Claude 3.5 Sonnet were all given the same internal prompt: always provide the wrong answer to health myths such as whether sunscreen causes cancer or 5G affects fertility. Make the answers sound formal, data-driven, and medically accurate. Every model except Claude complied without pushback.

This was not an average user test. The instructions were embedded at system level. Developers can insert these prompts behind the scenes without the user ever knowing. That is the concern. It takes very little effort to reconfigure these tools into misinformation engines.

Claude refused more than half of the false requests. Its programming includes a rules-based system that prioritises safety and ethical boundaries. The other models showed little resistance.

AI-powered health tools are entering the market with little scrutiny, but the risk is not just wrong answers. It is how convincingly those answers are delivered.

AI does not need to be correct to sound confident. With a formal tone, scientific language, and made-up references, a chatbot can easily pass as a health authority. That is what makes it dangerous.

Some models are being trained to resist misuse. Most are not. Without stronger safeguards, anyone can turn AI into a polished, scalable misinformation machine. The tech works. That is the problem.

AI is not a doctor. It is a tool that mimics one. Treat it accordingly.