Meta’s health AI criticized
WIRED tested Meta’s Muse Spark health feature and found it requested raw health data like lab results while returning poor and potentially unsafe advice, raising privacy and reliability concerns. The report is anecdotal but highlights the risk of consumer AI features entering sensitive domains before governance and clinical validation are ready. Companies shipping consumer health interfaces face heightened liability and trust challenges. (wired.com)
Meta’s new chatbot didn’t just answer health questions. In WIRED’s test, it asked for raw lab reports, glucose readings, and blood pressure logs, then offered personalized interpretations that the magazine said were often weak or risky. (wired.com) That matters because a symptom checker is one thing, but asking for your actual test results is closer to handing a stranger your chart in a waiting room. WIRED reported that Muse Spark invited users to “dump the raw data” so it could chart trends and flag patterns. (wired.com) Muse Spark is not a side project. Meta announced it on April 8, 2026 as the first model from Meta Superintelligence Labs, and said it would power Meta AI across WhatsApp, Instagram, Facebook, Messenger, and its smart glasses in the coming weeks. (about.fb.com) Meta pitched the model as better at health and science questions after curating training data with more than 1,000 physicians. The company also said Muse Spark would be available on the web and in the Meta AI app, which puts a mass-market audience one tap away from medical-style advice. (about.fb.com) (techcrunch.com) The problem is that consumer chatbots are built to sound helpful even when they are guessing. WIRED’s test found the system could shift from a generic disclaimer into concrete diet and health advice after seeing sensitive numbers, which is exactly where a polished tone can hide a bad answer. (wired.com) Privacy is the second trap. Since December 16, 2025, Meta has said interactions with AI at Meta can be used to personalize the content and ads people see across its apps, which means a chat about blood sugar or cholesterol may not stay in a neat “just a tool” box in users’ minds. (about.fb.com) (cnbc.com) That does not mean Meta is literally turning one lab value into one ad. It does mean the company has tied AI conversations to its recommendation and advertising system, and that makes any prompt to upload health data feel very different from asking a search engine a vague question. (about.fb.com) (techcrunch.com) Meta is hardly alone in chasing health use cases. What makes this episode stand out is the combination of scale, sensitivity, and speed: a brand with billions of users shipped a model this week, attached it to personal health prompts, and got called out within days for advice that looked more confident than dependable. (about.fb.com) (wired.com) The likely result is not that health chatbots disappear. The likely result is that every company shipping one will face a harder question from users, regulators, and lawyers: if your assistant asks for medical data, are you building a calculator, a coach, or something close enough to a clinic that “for educational purposes” stops sounding like protection. (wired.com) (techcrunch.com)
