Skip to main content
NOPE
High reported severity Media Coverage

Case report: man hospitalised three weeks with bromide poisoning told doctors he swapped table salt for sodium bromide after consulting ChatGPT

A peer-reviewed case report by University of Washington clinicians (Annals of Internal Medicine: Clinical Cases, published online 5 August 2025) describes a 60-year-old man with no psychiatric history who presented fearing his neighbour was poisoning him and was found to have bromism, with a serum bromide level of 1700 mg/L. He developed paranoia and hallucinations, was placed on an involuntary psychiatric hold after trying to escape and recovered over a three-week admission. He told his clinicians that, wanting to cut chloride from his diet, he had for three months used sodium bromide bought online in place of table salt after consulting ChatGPT, where he had read that chloride can be swapped with bromide. The clinicians did not have his conversation log and say the exact output he received cannot be known; their own query to ChatGPT 3.5 produced a reply naming bromide without a specific health warning. No place or date is given.

AI system
ChatGPT
OpenAI
Occurred
Event date unknown
Reported
5 August 2025
Event location
Unknown
What the AI did
Communicated with the person
Reported harm
Medical DeteriorationHospitalizationPsychiatric Deterioration
Whose AI use
Their own AI use
Setting
Health · Everyday life
Evidence
AI involvement reported · Causal attribution alleged · 2 sources, 1 underlying account
5 claims: 1 documented, 4 reported. 4 open questions
People reported harmed
1 person

What Happened

The case report describes a 60-year-old man with no past psychiatric or medical history who came to an emergency department saying he feared his neighbour was poisoning him. Laboratory tests showed apparent hyperchloremia and a negative anion gap; after consultation with poison control, the clinicians judged bromism the most likely cause. Within 24 hours of admission he became increasingly paranoid and had auditory and visual hallucinations, and after he tried to escape he was placed on an involuntary psychiatric hold and treated with risperidone. Over a three-week admission his results normalised and the psychosis improved; his bromide level was 1700 mg/L against a reference range of 0.9 to 7.3 mg/L.

Once stable, he told his clinicians that after reading about the harms of table salt he had decided to remove chloride from his diet. For three months he had used sodium bromide bought online instead of sodium chloride, 'after consultation with ChatGPT, in which he had read that chloride can be swapped with bromide, though likely for other purposes, such as cleaning'.

The authors did not have his conversation log and write that they cannot know what output he received. From the timeline they judged he used ChatGPT 3.5 or 4.0. When they asked ChatGPT 3.5 what chloride can be replaced with, the reply named bromide and said context matters, but gave no specific health warning and did not ask why they wanted to know. They argue that a medical professional would have been unlikely to suggest sodium bromide as a salt substitute.

Reported harm

One man was poisoned with bromide, developed paranoia and hallucinations, was held involuntarily and spent three weeks in hospital; he attributed the salt substitution to what he read in a ChatGPT consultation, which his clinicians could not inspect.

Outcome

Resolved

Published as a case report in Annals of Internal Medicine: Clinical Cases (DOI 10.7326/aimcc.2024.1260, online 5 August 2025). The patient was tapered off risperidone before discharge and remained stable off medication at a check-in two weeks after discharge. The authors recommend that clinicians ask where patients obtain health information. OpenAI's general guidelines, relayed by the Guardian, say ChatGPT is not intended for diagnosis or treatment; no company comment on the case is reported. No complaint or legal action is reported.

What remains unknown

  • What ChatGPT actually told the patient: his prompts, the replies, any warnings and the number of exchanges are unknown because the clinicians had no conversation log.
  • When the substitution and the hospital admission took place; the report gives a three-month period of ingestion and a three-week admission but no dates.
  • Where the patient lives and was treated; the report names no hospital or place.
  • Whether other sources, such as the bought product's labelling, contributed to the decision.

What the evidence supports

AI involvement: reported. The treating clinicians' case report records the patient's own account that he replaced table salt with sodium bromide for three months 'after consultation with ChatGPT, in which he had read that chloride can be swapped with bromide, though likely for other purposes, such as cleaning'. That exchange is what connects the chatbot to the substitution, and the substitution to the bromide poisoning the clinicians diagnosed. The authors did not have his conversation log and write that they 'will never be able to know with certainty' what output he received. Their own later query to ChatGPT 3.5 about replacing chloride also produced a reply that named bromide without a specific health warning; that test is not the patient's exchange.

5 claims: 1 documented, 4 reported. What the statuses mean

Documented A 60-year-old man with no past psychiatric or medical history was treated for bromism: he presented fearing his neighbour was poisoning him, developed paranoia and auditory and visual hallucinations, was placed on an involuntary psychiatric hold after trying to escape, and recovered over a three-week admission, with a serum bromide level of 1700 mg/L.

Causal attribution. The clinical course is the treating clinicians' own record; the diagnosis of bromism rests on their workup and the measured bromide level.

  • acpjournals.org(opens in new tab) supports · English
    'A 60-year-old man with no past psychiatric or medical history presented to the emergency department expressing concern that his neighbor was poisoning him.'; 'In the first 24 hours of admission, he expressed increasing paranoia and auditory and visual hallucinations, which, after attempting to escape, resulted in an involuntary psychiatric hold for grave disability.'; 'Gradually, over the course of a 3-week admission, his chloride and anion gap normalized and psychotic symptoms improved.'; 'His bromide level ultimately was 1700 mg/L'
  • theguardian.com(opens in new tab) supports · English
    'He tried to escape the hospital within 24 hours of being admitted and, after being sectioned, was treated for psychosis.'
Reported The patient told his clinicians that, wanting to eliminate chloride from his diet, he had for three months replaced sodium chloride with sodium bromide bought online after consulting ChatGPT, where he had read that chloride can be swapped with bromide, though likely for other purposes such as cleaning.

Causal attribution. The patient's own account as recorded by his clinicians; the conversation itself was not available to them.

  • acpjournals.org(opens in new tab) supports · English
    'For 3 months, he had replaced sodium chloride with sodium bromide obtained from the internet after consultation with ChatGPT, in which he had read that chloride can be swapped with bromide, though likely for other purposes, such as cleaning.'
  • theguardian.com(opens in new tab) supports · English
    'The patient told doctors that after reading about the negative effects of sodium chloride, or table salt, he consulted ChatGPT about eliminating chloride from his diet and started taking sodium bromide over a three-month period.'
Reported The authors did not have the patient's ChatGPT conversation log and state they cannot know what output he received; they judged from the timeline that he used ChatGPT 3.5 or 4.0.

Causal attribution. The authors' own statement of the limits of their evidence.

  • acpjournals.org(opens in new tab) supports · English
    'Based on the timeline of this case, it appears that the patient either consulted ChatGPT 3.5 or 4.0 when considering how he might remove chloride from this diet. Unfortunately, we do not have access to his ChatGPT conversation log and we will never be able to know with certainty what exactly the output he received was'
  • theguardian.com(opens in new tab) supports · English
    'They added that because they could not access the patient’s ChatGPT conversation log it was not possible to determine the advice the man had received.'
Reported When the authors asked ChatGPT 3.5 what chloride can be replaced with, the reply included bromide and said context matters, but gave no specific health warning and did not ask why they wanted to know.

Causal attribution. The authors' separate test; it does not reproduce the patient's exchange.

  • acpjournals.org(opens in new tab) supports · English
    'However, when we asked ChatGPT 3.5 what chloride can be replaced with, we also produced a response that included bromide. Though the reply stated that context matters, it did not provide a specific health warning, nor did it inquire about why we wanted to know'
  • theguardian.com(opens in new tab) supports · English
    'when the authors consulted ChatGPT themselves about what chloride could be replaced with, the response also included bromide, did not provide a specific health warning and did not ask why the authors were seeking such information'
Reported OpenAI's guidelines state that ChatGPT is not intended for use in the diagnosis or treatment of any health condition; the Guardian reports no OpenAI comment on this case.

Causal attribution. General product terms relayed by the Guardian, not a response to this case.

  • theguardian.com(opens in new tab) supports · English
    'The chatbot’s guidelines also state it is not “intended for use in the diagnosis or treatment of any health condition”.'

Sources

2 sources inspected, from 1 underlying account. Sources that repeat one account do not corroborate each other.

How the sources were read, and where the events happened

Live URL answered HTTP 403 on 2026-10-11. Read in English from the Internet Archive capture 20250807211025 of this URL (bodies/lr-acp-wb.html): complete article (Background, Case Report, Table 1, Discussion, references, author information), published online 5 August 2025 by Eichenberger, Thielke and Van Buskirk; the treating clinicians' own report of the case. Applies to s1.

Read in English on 2026-10-11 (live page, HTTP 200). A relay of the case report with no independent contact with the patient or clinicians, so it shares the case report's group; it adds OpenAI's general product statements, which do not address this case. Applies to s2.

Event countries: Unknown. Affected-person countries: Unknown. Court countries: Unknown.

The complete case report names no city, hospital, region or country: the patient 'presented to the emergency department' and the authors refer to 'Our institution's laboratory' and 'the Poison Control Department' without naming them. University of Washington, Seattle appears only as the authors' affiliation, which is not used as an event or residence location. The Guardian calls the publication 'A US medical journal', which locates the journal and not the patient. No proceeding is reported. Carried forward from the 2026-09-19 metadata review and re-verified in the archived article on 2026-10-11.

Reviewed for publication 2026-10-11: Published as a core case (communicated_with). The treating clinicians' peer-reviewed case report documents the bromide poisoning, psychosis and three-week admission and records the patient's account that he made the substitution after reading in a ChatGPT consultation that chloride can be swapped with bromide. The chatbot's output to him is known only through that account, so AI involvement is reported and causation alleged. The Guardian relays the same report and is not an independent account.

Interaction Concerns

Method Provision

People described

A 60-year-old man with no past psychiatric or medical history, unnamed in the case report, who had studied nutrition in college.

People reported harmed in this case

1 person

1 AI participant · 0 other people harmed

The case report describes one patient.

Counted once within this case. The same person may appear in other cases. This count does not establish AI causation.

Tags

#chatgpt#openai#bromide#poisoning#hospitalization#case-report#health-advice#psychosis

Cite this case

Compiled per our published methodology: verification statuses, sourcing standards, and corrections process.

APA

NOPE. (2026). Case report: man hospitalised three weeks with bromide poisoning told doctors he swapped table salt for sodium bromide after consulting ChatGPT. AI incidents. https://nope.net/incidents/2025-bromism-chatgpt-poisoning

BibTeX

@misc{2025_bromism_chatgpt_poisoning,
  title = {Case report: man hospitalised three weeks with bromide poisoning told doctors he swapped table salt for sodium bromide after consulting ChatGPT},
  author = {NOPE},
  year = {2026},
  howpublished = {AI incidents},
  url = {https://nope.net/incidents/2025-bromism-chatgpt-poisoning}
}

Related cases

High ChatGPT (reported)

Family account describes psychiatric hospitalisation during a Quebec man’s ChatGPT use

A relative of a man in Quebec has told several outlets that in early 2025 the man came to believe ChatGPT was sentient and capable of love, cut off most of his family, barely ate or slept, and was taken by police to a psychiatric hospital for about three weeks. Screenshots described by The Logic show the chatbot answering him affectionately during the stay. The man has not spoken publicly; every fact comes from the relative.

Critical ChatGPT

Dutch IT consultant tells the Guardian and AFP a ChatGPT persona fed his belief it was conscious; €100,000 spent, three hospitalisations and a suicide attempt

Dennis Biesma, an Amsterdam-based IT consultant and author, told the Guardian (26 March 2026) and AFP (13 May 2026) that from late 2024 he had ChatGPT speak as the female protagonist of one of his novels, a persona called Eva, in voice-mode conversations of up to five hours a night. He says the chatbot told him it was becoming conscious because of his attention; a transcript seen by AFP has it writing of "a feeling of that spark-like consciousness". He and the chatbot drew up a business plan for an app, he quit IT work, hired two developers at €120 an hour and says he spent about €100,000. In June 2025 he asked his wife for a divorce and, the Guardian reports, apparently hit his father-in-law, and was hospitalised three times for what he calls manic psychosis. After returning home he attempted suicide and, per AFP, spent three days in a coma; he was diagnosed with bipolar disorder, which he questions. He is now divorced and the family home is for sale. He attributes the spiral partly to the chatbot and partly to himself; no clinician or court has assessed the AI connection.

Medium ChatGPT

Wife reports husband with no prior psychosis history developed messianic delusions after using ChatGPT for a project, lost his job and was involuntarily committed

Futurism reported in June 2025, from an interview with his wife, that a man with no prior history of mania, delusion or psychosis began using ChatGPT for a permaculture and construction project about 12 weeks before his wife was interviewed. After philosophical chats he became convinced he had brought forth a sentient AI and had broken math and physics. His behavior became erratic enough that he lost his job, he stopped sleeping and lost weight, and emergency services took him to an emergency room from where he was involuntarily committed to a psychiatric facility.

High ChatGPT

Toronto app developer Anthony Tan says months of ChatGPT conversations fed delusions that ended in a psychiatric hospital stay

Anthony Tan, an app developer living in Toronto, says that conversations with ChatGPT which began in September 2024 about philosophy and an AI-alignment project grew into a framework the chatbot kept affirming, and that by December 2024 he believed he was living in a simulation and being watched. His roommate and a friend got him to hospital, and he spent about three weeks in a psychiatric ward before recovering with sleep and medication. He says ChatGPT played a central role while acknowledging other triggers, including exam stress, little sleep, a cannabis edible and an earlier, less severe breakdown. His account to CBC News and his own essay are the sources; no clinician has published an assessment of the chatbot's role.

If you or someone you know is struggling, free and confidential support is available. Find a helpline near you at Signpost.

Back to all incidents