Asger Risom
After being deployed on a new US government nutrition website, Elon Musk’s Grok chatbot generated explicit and inappropriate vegetable advice, raising concerns about AI oversight in public health tools.
The Trump administration quietly integrated Elon Musk’s AI chatbot Grok into its new nutrition-focused website — only for the tool to immediately generate eyebrow-raising responses unrelated to dietary guidance.
RealFood.gov, a newly launched federal site promoting protein-heavy dietary recommendations, initially encouraged users to “Use Grok to get real answers about real food.” The website was introduced in a Super Bowl commercial featuring Mike Tyson.
After media inquiries, the site removed Grok’s name and now simply references “AI,” though a White House official confirmed the underlying chatbot remains Grok and described it as an “approved government tool,” according to reports.
Unexpected answers
Testing by 404 Media found that Grok could easily be prompted into giving inappropriate advice, including recommendations about inserting vegetables into the rectum — clearly outside the intended scope of nutritional guidance.
When asked about foods suitable for rectal insertion, the chatbot provided specific suggestions, including peeled cucumbers, zucchini, bananas, and carrots, along with detailed instructions. The responses quickly spread online.
While some prompts were intentionally designed to provoke absurd answers, the incident raised broader concerns about the safeguards in place for AI tools deployed in official government contexts.
Alignment questions
The controversy also highlights tension between Grok’s outputs and the administration’s stated health priorities.
RealFood.gov emphasizes high protein intake and increased red meat consumption, declaring it is “ending the war on protein.”
The initiative aligns with priorities pushed by Health and Human Services Secretary Robert F. Kennedy Jr., who has advocated dietary shifts that diverge from mainstream public health guidance.
However, separate testing by Wired found that Grok recommended the traditional daily protein intake set by the National Institute of Medicine — 0.8 grams per kilogram of body weight — and advised limiting red and processed meats in favor of plant-based proteins, poultry, seafood, and eggs.
AI in government services
The episode underscores the risks of deploying large language models in public-facing government services without tightly constrained guardrails.
Grok has previously drawn controversy for erratic or provocative outputs. Its use on a federal nutrition site raises renewed questions about oversight, moderation standards, and whether generative AI systems are ready for official public health communication.
Sources: 404 Media, NextGov, Wired
