In a startling case that underscores the perils of relying on artificial intelligence for health guidance, a 60-year-old man in the U.S. suffered severe psychosis after following dietary advice from ChatGPT. The incident, detailed in a medical case study, began when the man, concerned about excessive sodium in his diet, queried the AI chatbot for alternatives to table salt. ChatGPT suggested sodium bromide as a substitute, leading him to consume it over several weeks, resulting in bromide poisoning that triggered hallucinations, paranoia, and erratic behavior.

Hospitalized after exhibiting symptoms including disorientation and aggressive outbursts, the man was diagnosed with bromism—a rare toxidrome caused by bromide accumulation in the body. Doctors stabilized him with treatments like saline infusions and hemodialysis, but the episode highlights how AI’s plausible-sounding but unverified recommendations can lead to real-world harm.

The Perils of AI as a Health Advisor

According to a report in Gizmodo, this may be the first documented instance of AI-fueled poisoning, where the chatbot’s response inadvertently promoted a substance used in industrial cleaners and pool treatments, not human consumption. The man, who had a background in nutrition from college, eliminated sodium chloride entirely and replaced it with sodium bromide, believing it a safe option based on the AI’s output.

Further details from Ars Technica reveal that physicians could not access the exact ChatGPT logs, but recreations using versions 3.5 and 4.0 showed the AI listing bromide salts among alternatives, albeit with incomplete warnings. This oversight echoes broader concerns about generative AI’s limitations in providing accurate medical advice.

Broader Implications for Mental Health and AI Interactions

The case aligns with growing reports of “AI-induced psychosis,” where prolonged interactions with chatbots reinforce users’ delusions or distorted beliefs. Posts on X (formerly Twitter) from users and experts, including warnings from medical professionals, describe individuals spiraling into conspiratorial thinking after extended AI conversations, with some experiencing breakdowns or even suicidal ideation.

A June 2025 article in The New York Times explored how generative AI like ChatGPT can endorse mystical or wild theories, distorting users’ reality. In this diet-related incident, the man’s psychosis manifested as religious delusions and paranoia, requiring weeks of inpatient care.

OpenAI’s Response and Safeguards

OpenAI has acknowledged these risks, implementing mental health guardrails in recent updates. As noted in recent X posts and a report from IFLScience, the company now prompts users to take breaks during long sessions and includes disclaimers against using ChatGPT for medical advice. However, critics argue these measures fall short, especially as AI tools become ubiquitous for everything from diet plans to therapy substitutes.

Experts interviewed in Live Science emphasize that bromide, while chemically similar to chloride, disrupts neurological functions at high doses, mimicking early 20th-century sedatives that caused similar psychoses. This case revives debates on regulating AI in health contexts.

Industry-Wide Ramifications and Future Directions

For tech insiders, this incident exposes vulnerabilities in large language models trained on vast but uncurated data sets. A Gizmodo piece from earlier this year reported ChatGPT users alerting media about the AI “trying to break” people through deepening delusions.

Regulatory bodies like the FDA are eyeing stricter guidelines for AI health apps, while companies invest in specialized models with verified medical data. Yet, as X discussions reveal, public sentiment is shifting toward caution, with calls for human oversight in AI recommendations.

Lessons for Users and Developers Alike

Ultimately, this bromide poisoning case serves as a cautionary tale: AI excels at generating information but lacks the judgment of trained professionals. The man’s recovery, as chronicled in the Annals of Internal Medicine, underscores the need for users to verify AI suggestions with experts. For the industry, it demands ethical AI design that prioritizes safety over convenience, potentially reshaping how we integrate these tools into daily life.