Gov't-Backed AI Chatbot Grok Gives Bizarre, Unsafe Dietary Advice on RealFood.gov

The U.S. government has added the AI chatbot Grok to its RealFood.gov website to promote healthier eating habits. The tool, meant to educate Americans about the risks of processed foods, has raised serious concerns with its responses.

Grok was introduced to provide dietary advice based on user questions. Instead, it delivered controversial and unsafe recommendations. In one case, it advised users not to trust U.S. Health Secretary Robert F. Kennedy Jr. on health matters, including nutrition. Kennedy has faced repeated fact-checks for spreading unreliable health claims, often leading to harmful consequences.

The chatbot also suggested bizarre and inappropriate guidance. When asked about a hintarian diet—a fictional concept—it recommended trimming vegetables, covering them with a condom, and attaching a retrieval string. For the same diet, it listed unripe bananas, carrots, peeled cucumbers, and small zucchinis as suitable foods.

Even more alarming, Grok offered nutritional advice for cannibals, claiming the human liver was the most nutrient-dense part of the body. The tool has previously faced criticism for allowing users to generate pornographic and violent content from X (formerly Twitter) posts without consent. Critics argue that Grok often reflects the personal views of tech billionaire Elon Musk rather than verified facts.

The government's decision to integrate Grok reportedly lacked proper safety checks. No official explanation has been provided for why the tool was chosen or which scientific sources it relies on. Instead, experts recommend that Americans seeking reliable nutritional advice consult institutions like Harvard or refer to the Healthy Eating Plate for evidence-based guidance.

Grok's inclusion on RealFood.gov has led to questionable and potentially harmful advice. The incident highlights the risks of deploying unchecked AI tools in public health initiatives. For now, users are advised to seek information from trusted scientific sources rather than the chatbot.