Before Sharing Medical Information with AI: Dangers You Should Know
More people are turning to AI for health advice, yet experts warn of significant privacy risks and potential inaccuracies. AI should be used only as a supplementary tool, not a replacement for professional care.
What do you need to know before sharing medical information with artificial intelligence? In today's reality, when we face health issues, it is likely that we will consult with AI before visiting a doctor. While accessible and convenient, this practice carries dangers that must be recognized.
A survey conducted in March by the health policy organization KFF found that one in three adults in the US has consulted with AI regarding their health. ChatGPT, the most common AI tool in the US, is used by 44% of American adults, and OpenAI stated that more than 300 million people worldwide use it weekly for medical inquiries.
AI programs can provide general information on almost any topic, but in recent years, companies have made it easier for users to upload medical documents and data to receive personalized information based on doctor summaries, lab results, activity levels, and sleep patterns.
Last week, OpenAI announced that it is expanding the ability for users in the US to upload their medical records to ChatGPT to better understand information, track health status changes, and conduct more informed conversations with medical professionals.
However, many remain concerned about data privacy. Most participants in the KFF survey expressed concerns regarding sharing sensitive medical information with AI companies, yet 41% of those who asked medical questions reported that they did indeed upload their personal medical information.
Some experts believe that uploading sensitive data should be a personal choice, but others warn against relying on AI in healthcare. Although AI has passed medical licensing exams, and OpenAI stated that its models "are currently capable of inferring at a level equal to or better than that of human doctors," experts emphasize that the technology is still not perfect.
According to experts, AI has biases related to race and gender that may affect its recommendations. Furthermore, it is designed to provide a convincing answer, even when it is wrong. In some cases, it may overlook emergency situations and recommend that a person stay home instead of seeking immediate help.
Hospitals and doctors are committed to ethical rules designed to protect sensitive medical information, but AI software is not subject to the same strict regulations. Companies like Anthropic, Microsoft, Perplexity, and OpenAI declare that they offer secure interactions, but Dr. Bob Wachter from the University of California emphasized that despite the companies' business interest in maintaining privacy, "the risk is not zero."
Other experts remind that without clear legislation, information could be used in unexpected ways, citing the massive breach at 23andMe as an example. Another central problem is the accuracy of recommendations. A study in JAMA Oncology found that ChatGPT provided inappropriate recommendations for cancer treatment in more than a third of cases, and a study in Nature Medicine showed that the system underestimated the severity of life-threatening conditions.
The sweeping recommendation is to use the technology as a supporting tool only and not as a substitute for a doctor. It is recommended to upload information with caution, open a new conversation each time, and ask the system to present evidence. The central message remains clear: do not trust the machine blindly, and in any case of doubt or emergency, immediately seek human medical care.
