He only wanted to cut down on salt. A 60-year-old, ordinary in every way, read the usual warnings about sodium and decided to be proactive. Like millions of the rest of us, he opened a chat window and asked an AI for ideas. What followed sounds like a parable for the internet age: he swapped table salt for sodium bromide, yes, the industrial chemical, and over three months drifted into paranoia, insomnia, rashes, and hallucinations. Doctors finally named it: bromism, a 19th-century toxicity making a 21st-century comeback through a screen.
The clinical facts are chilling precisely because they are mundane. He was not chasing miracle cures or downing mystery powders; he followed a substitution that looked “chemically adjacent,” then quietly unraveled. Physicians say he arrived convinced a neighbor was poisoning him. Labs and time told a simpler story: he had been poisoning himself. The kicker? When researchers tried similar prompts, they say the chatbot suggested bromide as a “swap” for chloride with inadequate warnings. Even without his chat logs, the recreation underscores how AI can surface context-free answers that sound authoritative but ignore the messy, embodied reality called a human body.
Related article - Uphorial Podcast
But this is not just a gotcha against AI; it is a mirror held up to us. We already outsource memory to phones, wayfinding to maps, and taste to algorithmic feeds. Health, though, is painfully non-abstract. Sodium chloride is not just “salt,” it is physiology: nerves firing, fluids balancing, cells maintaining electrical gradients. Replace it with bromide, and the body misreads signals. The mind blurs. A metaphor becomes literal; you cannot prompt engineer your way out of biochemistry. Bromide once lived in pharmacies and is now found more in pool chemicals than pantries. That distinction matters; a probabilistic model can collapse it.
So what do we think about it? First, we should resist cartoonish blame. A chatbot did not climb into his kitchen. Yet we also should not let techno optimism sand down the edges. When advice concerns food, drugs, or mental health, “good enough” is not good enough. Doctors called it “highly unlikely” that a clinician would recommend bromide as a culinary substitute; that gap is the point. AI predicts plausible text, not consequences. It does not taste like the soup it is seasoning.
Second, the context is not a one-off scare. Research and audits keep finding that chatbots can produce risky or harmful guidance, especially for vulnerable users, unless systems and guardrails are deliberately designed for safety and constantly updated. If you are going to ask AI about your body, your mind, or your meds, treat it like a searchlight, not a compass. Use it to illuminate questions; let clinicians and verified sources set direction.
Third, there is a human undercurrent here worth lingering on: the man was trying to be healthier. That is the quietly tragic center of the story. Health anxiety meets infinite answers; caution meets confidence theater. The lesson is not “never use AI.” It is “slow down.” Before you change something you eat every single day, triangulate: check a reputable medical site, call your clinician, ask a pharmacist. If the suggestion sounds unusual, assume it is.
Where do we go from here? On the developer side: safer defaults, context-aware refusals, health-specific rails, and explicit citations. On the media side: less sensational framing, more biochemical clarity. On our side, the side with bodies, humility. Ask better questions. Demand sources. Treat AI answers like a draft, not a diagnosis. Because the saddest thing about this case is not the algorithm, it is that a man seeking a little less salt found a lot more harm, and all it might have taken to prevent it was one more question asked of the right person.
Note: If you are considering any diet change, consult a licensed clinician or registered dietitian; do not rely on AI for medical advice.
