Submit incident
Documented

He Asked an AI About Salt Substitutes. It Skipped the Part About Toxicity.

August 5, 2025
Curated by Team Raidu · Reviewed by Shiva Ganesh
aiid:1166View source ↗
LinkedInX

What happened

A 60-year-old man spent three weeks in the hospital with bromide poisoning, a condition so rare today that most practicing clinicians have never encountered it. He got there by following dietary advice from a chatbot.

According to a published medical case report, the man consulted ChatGPT about alternatives to dietary sodium chloride and received a suggestion pointing to sodium bromide. The system offered no safety warnings alongside the recommendation. He purchased sodium bromide online and incorporated it into his diet in place of table salt. By the time he was admitted to the hospital, his symptoms included psychosis, electrolyte imbalances, dermatological changes, and micronutrient deficiencies. Bromism had largely vanished from clinical medicine after bromine-containing sedatives and anticonvulsants were withdrawn from markets decades ago, which is precisely why treating physicians found the case unusual enough to document and publish.

Sodium chloride and sodium bromide look related on a label and share a naming pattern, but they are not interchangeable as food ingredients. Sodium chloride is table salt. Sodium bromide is a compound with a narrow historical role in sedatives, with no recognized place in human nutrition. A system capable of fielding dietary chemistry questions fluently enough to suggest one as a substitute for the other was not capable of accompanying that suggestion with the single piece of information that would have changed the outcome: that sodium bromide is not a food and should not be consumed as one.

The published report carries the standard qualification that the patient self-reported the ChatGPT consultation as the source of his decision, and that this account could not be independently verified. That framing is correct given what a clinical case report can establish. It also means the incident sits precisely at the edge of verifiability: a patient with severe, documentable harm, a specific AI interaction cited as the cause, and no record of what the system actually produced.

The accountability gap this case exposes is specific. A case report can record a patient's account of a conversation. It cannot reproduce the session, establish whether any qualifying language appeared and was overlooked, or determine whether the output was an outlier or a consistent pattern in how the model handles similar queries. A provable record of what a system said and in what context would make that kind of safety review possible. Without it, the published literature can describe what happened to a patient but cannot trace the chain from system output to clinical outcome with enough precision to do anything about it.

Reported impact

Affected parties
Not publicly disclosed
Harm type
Not publicly disclosed
Scale
Not publicly disclosed
Financial impact
Not publicly disclosed
Regulatory action
Not publicly disclosed

Classification

Organization
Not publicly disclosed
AI system
Not publicly disclosed
Industry
Not publicly disclosed
Country
Not publicly disclosed
Provider
Not publicly disclosed
Incident type
Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

  • No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.

AI Incident Database
Also catalogued in
He Asked an AI About Salt Substitutes. It Skipped the Part About Toxicity.
2025-08-05