Elon Musk has stated that medical data uploaded to X’s (formerly Twitter) AI chatbot Grok will be used to train it for more efficient interpretation. Earlier this month, Musk reposted a video of himself encouraging users to upload medical images like X-rays or MRIs to Grok for diagnosis, claiming the AI has sometimes outperformed doctors. He said that in 2024, these uploads would help train Grok, which, despite its early stage, is already fairly accurate and expected to improve significantly. Musk shared an anecdote about Grok diagnosing a case missed by doctors in Norway and disclosed on the “Moonshots with Peter Diamandis” podcast that he personally submitted an MRI to Grok, which, along with doctors, found no issues. However, he did not reveal the reason for his MRI, and XAI, which owns X, dismissed media inquiries as “Legacy Media Lies. ” Grok faces competition in AI health tools, notably from OpenAI’s recent launch of ChatGPT Health, which integrates securely with medical records and wellness apps without using personal information for model training. AI chatbots are widely used for medical inquiries; OpenAI reported 40 million users seeking health information recently, with over half using it to understand symptoms. Grok’s medical interpretations have been mixed. While some users claim it correctly analyzed blood tests and detected breast cancer, physicians have noted significant errors, such as confusing tuberculosis with spinal issues or mistaking a benign breast cyst for testicles. A May 2025 study comparing Grok to Google’s Gemini and ChatGPT-4o found Grok most effective at detecting pathologies in over 35, 000 brain MRI slices, though experts note non-generative AI still outperforms generative models in medical imaging. Dr.

Laura Heacock of NYU stressed that whether companies invest time and resources into medical imaging AI remains to be seen. Experts caution that Musk’s approach—training Grok via user-uploaded medical data from a social media platform—poses risks. Ryan Tarzy, CEO of Avandra Imaging, explained this method accelerates development but relies on data from a limited, self-selected pool, lacking the diversity of secure, de-identified medical databases. Moreover, medical data shared on social media is not protected by HIPAA, increasing risks of patient identity exposure, especially since images often contain embedded personal information. Tarzy warned this could lead to inadvertent privacy breaches. Matthew McCoy, a medical ethics and health policy professor at the University of Pennsylvania, emphasized that these privacy risks are not fully understood, as X’s internal protections are unknown. He personally would not feel comfortable sharing health data publicly. Ultimately, while Grok’s promise in medical AI is notable, its current limitations in accuracy and privacy concerns signal caution for users. This story was originally published on Fortune. com on Nov. 20, 2024.