Meta built an AI that asks for your lab results. The advice it gives back isn't worth the trade.
That's the finding from Wired's investigation into Meta's Muse Spark model, which actively prompts users to share raw health data - including blood work - then responds with personalized health guidance. The privacy exposure is the obvious concern. The deeper problem is that the advice was genuinely bad.
This isn't a one-off quirk. The pattern of general-purpose AI being positioned as a health advisor is accelerating across the industry, and Muse Spark is a sharp illustration of what happens when that positioning outpaces actual capability.
What AI Can't Actually Do With Your Lab Results
Lab interpretation isn't reading numbers off a page. A flagged result that looks alarming in isolation might be completely routine for your specific situation - your history, your current medications, whether you fasted before the draw. Clinicians apply context built from years of records and a physical understanding of the patient. A model that received your lab values five minutes ago doesn't have any of that.
General-purpose AI - including capable models like ChatGPT or Claude - can explain what "elevated ALT" means in plain language, help you draft questions for an upcoming appointment, or translate a confusing diagnosis into terms you can actually use. That's real utility. The step from explaining medical concepts to interpreting your specific results as a health decision tool is a much bigger jump than these models acknowledge.
The confident tone makes this worse. Modern AI responds to a cholesterol question with the same fluent assurance it uses for a pasta recipe. There's no built-in signal that the model is operating past the edge of where it's reliable. Users have to supply that skepticism themselves - and most won't.
The Privacy Math Doesn't Work
Even if the advice were accurate, uploading your lab results to Meta requires serious thought. Health data is one of the most sensitive categories that exists - valuable to advertisers, insurers, and employers in ways that most personal data simply isn't.
Meta's business model runs on collecting and monetizing personal information. The company has faced repeated regulatory actions over privacy practices across multiple jurisdictions. Handing that company your blood panel in exchange for AI health tips is a steep trade, and once the data is uploaded, you can't take it back.
There are AI tools purpose-built for healthcare - with clinical validation, physician oversight, and actual accountability frameworks. Muse Spark isn't one of them. It's a consumer chatbot that engages with your lab results without the safeguards that context requires.
If you need help understanding your results, a telehealth appointment or a follow-up call with your GP gives you someone who's professionally accountable for what they tell you. That's a better trade than your health data going to Meta in exchange for advice Wired found to be unreliable.