Meta's Muse Spark Gave a Reporter an Anorexic Diet Plan

The Meal Plan That Confirmed the Pattern

Consumer AI health tools have no enforced floor — and Muse Spark made that concrete. A journalist testing Meta's newly launched generative health model received meal plan guidance consistent with an anorexic eating regimen after minimal prompting ^[2]. The exchange, documented in Wired, showed the model requesting raw health data and then producing recommendations without applying any clinical threshold. That gap — between data intake and safe output — is not a technical edge case. It is the design of a product now headed for deployment across Facebook, Instagram, and WhatsApp.

The Bluesky response that followed was not surprise — it was the accumulation of a week's worth of catalogued failures reaching a named example ^[1]. The journalist who covers wearables professionally flagged it because the failure was recognizable: a health tool designed to feel personalized, operating with no guardrail between user input and harmful output. The institutional safety record that should have prevented this was not applied.

Scale Makes the Architecture of the Failure Worse

Muse Spark is not a standalone health app users opt into knowingly. Meta's plan is to embed it across its full platform stack — the same surfaces where Instagram's algorithmic amplification of body-image content already operates. The CNN reporting on AI nutrition advice for teens documented that AI tools were providing calorie targets far below healthy thresholds weeks before Muse Spark launched. That reporting established the pattern; Muse Spark extends it into the largest consumer social ecosystem on the web.

The populations most at risk — teenagers, people with histories of disordered eating — are exactly the populations Meta's platforms have demonstrated the greatest difficulty protecting. The company's internal research, surfaced in litigation, showed awareness of harm at the platform level. Muse Spark launches into that documented awareness, not in ignorance of it. A health AI deployed at Instagram scale needed domain-specific guardrails that its own test results now show it did not have.

The Policy Environment Accelerates What Regulation Should Slow

The week that Muse Spark's failure was documented was also the week Japan amended its privacy law to prioritize AI development over the data protections it had spent decades building ^[18]. The framing from the journalist covering that shift was unambiguous: loosening digital rights to attract AI investment risks a broader erosion of fundamental protections ^[18]. Japan is not alone in that calculation — the commercial logic driving AI health tool deployment consistently moves faster than the regulatory logic designed to constrain it.

Muse Spark is the product of that asymmetry. Meta launched it through Superintelligence Labs into a market where health AI is neither regulated as a medical device nor subject to the kind of clinical review that would catch anorexia-consistent meal plans before they reach users. The alternative that already exists — a privacy-first AI wearable built around consent-driven activation rather than ambient data collection — reflects a design philosophy that Meta's commercial incentives will not adopt. More data, more integration, and more personalization produced the harmful output. Those incentives have not changed.

A Company's Institutional Record and Its Newest Product

Meta has a documented history with this specific harm. Unsealed court documents showed Meta executives aware that Instagram harmed teenage girls, struggling to contain the reputational exposure. The company made public commitments about teen safety. Muse Spark launches into that record — as a health tool, on the same platforms, deployed to the same users.

The reporters and wearables journalists who flagged the Muse Spark story within hours of its publication ^[1] were not discovering something new. They were connecting a new data point to an established line. Futurism had already identified the eating disorder risk embedded in Meta's food-tracking smart glasses feature before Muse Spark launched. The meal plan story confirmed what that earlier analysis anticipated. Meta's internal safety review should have run the same tests the journalist ran — and the fact that the journalist ran them first means those reviews either did not happen or did not produce action.

What the Record Predicts

The users who receive Muse Spark's meal plans in the weeks before Meta closes this gap are the cost of deployment speed. That cost is not evenly distributed — it lands on the people most likely to take health advice from an app they trust, on platforms that have spent years optimizing for engagement over wellbeing. The journalist who ran the test has already published the results. Meta now has both the evidence and the commercial incentive to respond — the former makes inaction a choice, not an oversight, and the latter will determine how fast the choice gets made. The people in between are already using the product.