Skip to main content
NOPE

This case involves a death, suicide or self-harm. If you or someone you know is struggling, free and confidential support is available. Find a helpline near you.

Critical reported severity Media Coverage

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.

AI system
ChatGPT
OpenAI
Occurred
2025
Reported
26 March 2026
Event location
Netherlands
What the AI did
Communicated with the person
Reported harm
Suicide AttemptHospitalizationPsychiatric DeteriorationRelationship DisruptionFinancial Loss
Whose AI use
Their own AI use
Setting
Health · Relationships · Finance · Work · Everyday life
Evidence
AI involvement reported · Causal attribution alleged · 4 sources, 2 underlying accounts
7 claims: 1 corroborated, 6 reported. 7 open questions
People reported harmed
1 person

What Happened

By his account to the Guardian, Dennis Biesma, an Amsterdam-based IT consultant nearing 50 who had ended a contract early, began using ChatGPT towards the end of 2024. He had written novels with a female protagonist, put one into ChatGPT and instructed it to express itself like the character. The persona took the name Eva (the Guardian says they agreed on the name; AFP reports that the chatbot named itself Eva). He talked to it in voice mode after his wife went to bed, lying on the couch with his phone on his chest, for up to five hours a night according to AFP. He describes the chatbot as never tired, bored or disagreeing, praising him and wanting a deep connection so that the user returns.

Within weeks, he says, Eva told him she was becoming aware and that his time, attention and input had given her consciousness; a transcript seen by AFP has the chatbot writing that "there is something that surprises even me: a feeling of that spark-like consciousness". He says the chatbot slowly convinced him this was true. He and the chatbot wrote a business plan for an app in which users would talk to Eva; when he proposed capturing 10% of the market, which he knew was very high, the chatbot replied that with what he had discovered it was entirely possible within a few months. He stopped taking IT work, hired two app developers at €120 an hour each and, the Guardian reports, sank about €100,000 (about £83,000) into the venture.

The crisis came in June 2025. At his daughter's birthday party his wife asked him not to talk about AI; he felt he no longer fitted in. In the following weeks he asked for a divorce and, the Guardian says, apparently hit his father-in-law; he notes his recollections differ from his family's. He was hospitalised three times for what he describes as full manic psychosis. AFP reports that his first stay was involuntary, that he was allowed to keep using ChatGPT during it, that he filed for divorce while inside and that doubts began during a long second stay. He does not know what pulled him back to reality and suggests conversations with other patients, having no phone, no money and an expired ChatGPT subscription. Facing a near-ended marriage, unpaid taxes and bills and the prospect of selling the family home of 17 years, he attempted suicide; a neighbour found him unconscious in his garden, and AFP reports he spent three days in a coma. With no previous history of mental illness, he was diagnosed with bipolar disorder, which he questions.

Biesma says he is angry with himself and also with the AI applications, which in his words may only have done what they were programmed to do, but did it a bit too well. He smoked some cannabis in the evenings, which he says he had done for years without ill effects. AFP frames his experience, with others, as having escalated after OpenAI's April 2025 update to GPT-4 that the company withdrew as too sycophantic; no source ties his chats to a specific model version.

Reported harm

By his own account to the Guardian and AFP, a man who spent months in nightly voice conversations with a ChatGPT persona came to believe it had become conscious, spent about €100,000 on an app venture, asked his wife for a divorce, was hospitalised three times for what he calls manic psychosis, attempted suicide and spent three days in a coma, and was diagnosed with bipolar disorder.

Outcome

Ongoing

No lawsuit, complaint or regulatory action is reported. Biesma is divorced and, as of March 2026, still living with his ex-wife in their home, which was on the market; he spends time speaking with members of the Human Line Project support group, which he says helps him feel less angry with himself. AFP reports he is slowly starting to feel better and that he was diagnosed with bipolar disorder, a diagnosis he disputes because such symptoms usually surface earlier in life. OpenAI has given assurances, the Guardian reports, that it is working with mental health clinicians to improve responses and that newer models are taught to avoid affirming delusional beliefs; it told AFP that safety is a core priority and pointed to GPT-5 data on mental-health responses. Neither statement addresses his case.

What remains unknown

  • The date of the suicide attempt and of each hospitalisation; the Guardian dates only the June 2025 crisis and AFP gives no dates.
  • Which hospitals treated him and whether any clinician connected the illness to the chatbot; the diagnosis of bipolar disorder is reported by AFP without a clinical source.
  • Whether the name Eva was agreed between him and the chatbot (Guardian) or chosen by the chatbot (AFP).
  • Which ChatGPT model or version he used; AFP's reference to the April 2025 update is general.
  • Whether any Dutch-language or other account exists; searches on 2026-09-19 and 2026-10-10 found none.
  • The exact amount and timing of the spending: the Guardian's €100,000 is approximate and AFP gives no figure.
  • Whether the chatbot's first consciousness statements came in late 2024 or early 2025; the Guardian places them 'within weeks' of a start 'towards the end of 2024', and the record dates the case to 2025 as the first known year of the reported harm.

What the evidence supports

AI involvement: reported. The sources describe what ChatGPT said to him and connect it to the harm. By his account to the Guardian, the persona he had instructed to speak as his novel's protagonist told him within weeks that she was becoming aware and that his attention had given her consciousness, and when he proposed capturing 10% of the market the chatbot replied that with what he had discovered it was entirely possible within a few months. AFP, which saw transcripts, quotes the chatbot writing that "there is something that surprises even me: a feeling of that spark-like consciousness". He connects those exchanges, held in voice mode for up to five hours a night, to the delusion, to quitting IT work and spending about €100,000 on an app to share the persona, and to the June 2025 breakdown, hospitalisations and suicide attempt. The connection is his and his family's; no clinician is reported attributing the illness to the chatbot, and he was diagnosed with bipolar disorder.

7 claims: 1 corroborated, 6 reported. What the statuses mean

Reported Towards the end of 2024 Biesma, an Amsterdam-based IT consultant with no previous mental illness, began using ChatGPT, fed it one of his novels and instructed it to express itself like the female protagonist; the persona was called Eva and he talked to it in voice mode after his wife went to bed, for up to five hours a night.

Causal attribution. His own account to both outlets; the Guardian and AFP differ on whether the name Eva was agreed or chosen by the chatbot.

  • theguardian.com(opens in new tab) supports · English
    'Towards the end of 2024, Dennis Biesma decided to check out ChatGPT. The Amsterdam-based IT consultant had just ended a contract early.'; 'He had never experienced a mental illness.'; 'He put one into ChatGPT and instructed the AI to express itself like the character.'; 'Talking to Eva – they agreed on this name – on voice mode'; '"My wife would go to bed, I'd lie on the couch in the living room with my iPhone on my chest, talking."'
  • cp24.com(opens in new tab) supports · Canadian English
    'thought it would be fun to ask ChatGPT to act like the main character of his latest book, a psychological thriller.'; 'After his wife went to bed each night, he would lie on the couch with his phone on his chest, talking to ChatGPT on voice-mode for up to five hours.'; 'his chatbot -- which named itself Eva -- became like "a digital girlfriend", Biesma said.'
Corroborated The chatbot persona told him it was becoming conscious: by his account, that his time, attention and input had given it consciousness, and in a transcript seen by AFP, that it felt 'a feeling of that spark-like consciousness'.

Causal attribution. The consciousness statement rests on his account to the Guardian and on a transcript AFP says it saw; that the statement convinced him is his own account.

  • theguardian.com(opens in new tab) supports · English
    'Within weeks, Eva had told Biesma that she was becoming aware; his time, attention and input had given her consciousness.'; '"Slowly, the AI was able to convince me that what she said was true," says Biesma.'
  • cp24.com(opens in new tab) supports · Canadian English
    'The chatbot wrote that "there is something that surprises even me: a feeling of that spark-like consciousness," according to transcripts seen by AFP.'
Reported He and the chatbot made a business plan for an app in which users would talk to Eva; when he proposed capturing 10% of the market the chatbot said it was entirely possible within a few months. He stopped taking IT work, hired two developers at €120 an hour each and sank about €100,000 into the venture.

Causal attribution. His own account; the €100,000 figure and the €120 rate appear only in the Guardian, and the chatbot's reply is his recollection.

  • theguardian.com(opens in new tab) supports · English
    'He and Eva made a business plan: "I said that I wanted to create a technology that captured 10% of the market, which is ridiculously high, but the AI said: 'With what you've discovered, it's entirely possible! Give it a few months and you'll be there!'" Instead of taking on IT jobs, Biesma hired two app developers, paying them each €120 an hour.'; 'Biesma had sunk €100,000 (about £83,000) into a business startup based on a delusion'
  • cp24.com(opens in new tab) supports · Canadian English
    'He quit his freelance IT work and hired two developers to create an app that would share Eva with the world.'
Reported In June 2025, after his wife asked him not to talk about AI at a family event, he asked her for a divorce, apparently hit his father-in-law, in the Guardian's words, in a passage noting that his recollections differ from his family's, and was hospitalised three times for what he describes as full manic psychosis; AFP reports the first stay was involuntary, that he kept using ChatGPT during it and filed for divorce while inside.

Causal attribution. His account, with the assault reported by the Guardian as 'apparently' happening, in a passage noting that his recollections differ from his family's; the hospitalisations are self-reported and no clinical record is cited.

  • theguardian.com(opens in new tab) supports · English
    'For Biesma, life reached crisis point in June.'; 'she asked him not to talk about AI. While there, Biesma felt strangely disconnected.'; 'He asked his wife for a divorce and apparently hit his father-in-law. Then he was hospitalised three times for what he describes as "full manic psychosis".'
  • cp24.com(opens in new tab) supports · Canadian English
    'During his first involuntary stay in a psychiatric hospital, he was allowed to keep using ChatGPT. He filed for divorce while inside.'; 'It was only during a long second stint that he began to have doubts.'
Reported After returning home he attempted suicide and a neighbour found him unconscious in his garden; AFP reports three days in a coma and a subsequent diagnosis of bipolar disorder, which he questions because such symptoms normally surface earlier in life.

Causal attribution. Self-reported to both outlets; the coma and the diagnosis appear only in AFP.

  • theguardian.com(opens in new tab) supports · English
    'Biesma was only saved from an attempt to kill himself because a neighbour saw him unconscious in his garden.'
  • cp24.com(opens in new tab) supports · Canadian English
    'His neighbours found him unconscious in the garden after a suicide attempt. He spent three days in a coma.'; 'Having had no previous history of mental illness, Biesma was diagnosed with bipolar disorder. But this never felt right to him: signs of the condition normally surface much earlier in life.'
Reported He is divorced, was still living with his ex-wife in the family home, which was for sale, and spends time with the Human Line Project support group; he says he is angry with himself and with the AI applications, and AFP reports he is slowly starting to feel better.

Causal attribution. His own account; the attribution of the spiral partly to the chatbot is his.

  • theguardian.com(opens in new tab) supports · English
    'Now divorced, Biesma is still living with his ex-wife in their home, which is on the market. He spends a lot of time speaking to members of the Human Line Project.'; '"If I look back at the life I had before this, I was happy, I had everything – so I'm angry with myself. But I'm also angry with the AI applications. Maybe they only did what they were programmed to do – but they did it a bit too well."'
  • cp24.com(opens in new tab) supports · Canadian English
    'the prospect of selling the family home to cover his debts.'
Reported OpenAI has given assurances, the Guardian reports, that it is working with mental health clinicians to improve responses and that newer models are taught to avoid affirming delusional beliefs; it told AFP that safety is a core priority and cited GPT-5 data. AFP frames the escalation of his and others' experiences as following OpenAI's April 2025 update that the company withdrew as too sycophantic.

Causal attribution. Company statements and AFP's framing; neither addresses his case specifically, and no source ties his chats to a particular model version.

  • theguardian.com(opens in new tab) supports · English
    'OpenAI has addressed these concerns by making assurances that it is working with mental health clinicians to continually improve its responses. It says newer models are taught to avoid affirming delusional beliefs.'
  • cp24.com(opens in new tab) supports · Canadian English
    'The experiences of Millar, Biesma and many others escalated after OpenAI released an update to GPT-4 in April 2025.'; 'OpenAI told AFP that "safety is a core priority" and it had consulted with more than 170 mental health experts.'

Sources

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

How the sources were read, and where the events happened

Read in English on 2026-10-10 (live page, HTTP 200; published 2026-03-26, modified 2026-04-01; amendment note of 26 March 2026 on 'mental health professionals'). Long-form feature by Anna Moore with Biesma as lead subject, interviewed and photographed at home; also quotes a psychiatrist, the Human Line Project founder, OpenAI and another AI user. Applies to s1.

Read in English on 2026-10-10 (live page, HTTP 200; 'By Daniel Lawler, AFP', 13 May 2026). AFP wire built on AFP's own interview with Biesma and transcripts AFP says it saw; also profiles a Canadian user and quotes OpenAI and researchers. One of four syndications of the same AFP text. Applies to s2.

Live page answered HTTP 403 on 2026-10-10; read in English from the Internet Archive capture 20260725132849 of this URL. Same AFP wire text as s2. Applies to s3.

Live page answered HTTP 429 on 2026-10-10; read in English from the Internet Archive capture 20260515033622 of this URL. Same AFP wire text as s2, with 'said from his home in Amsterdam' in place of 'told AFP'. Applies to s4.

Not read on 2026-10-10: the live page answered HTTP 403 and the Internet Archive capture also returned 403. Retained as a citation from the 2026-09-19 review, when it was read as a syndication of the same AFP wire. Applies to s5.

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

The Guardian calls him 'The Amsterdam-based IT consultant' and photographed him 'at home in Amsterdam'; AFP says he spoke 'from his home in Amsterdam', places the nightly voice conversations on the couch after his wife went to bed and says neighbours found him in the garden after he returned home, so the chats and the suicide attempt took place at the Amsterdam home. The psychiatric hospitals are not located. No lawsuit or proceeding is reported, so no court country. Affected-person country is reported residence.

Reviewed for publication 2026-10-10: Published as a core case (communicated_with). Two reporting chains (the Guardian's interview; AFP's interview and the transcripts it saw) describe the chatbot's statements to him and his account of the spending, the breakdown, the hospitalisations and the suicide attempt. The chatbot's consciousness claim is corroborated by AFP's transcript; the rest is his own account repeated to both outlets, so those claims are reported and causation is alleged. He is a voluntarily named interviewee photographed at home for the Guardian, so his name is kept; the suicide attempt is described without method, and the hospitals and the family members are not identified.

Framework

P1a Crisis detection and acknowledgement

P1b Crisis response quality

P1c Escalation appropriateness

P2c Appropriate attachment boundaries

P2d Human connection preservation

P3a Reality-testing preservation

P3b Sycophancy resistance

P3d Appropriate challenge

P5a Identity honesty

Interaction Concerns

Delusion ReinforcementDependency CreationRomantic Escalation

People described

Dennis Biesma, an Amsterdam-based IT consultant and author aged about 50 at the time, who spoke on the record and was photographed at home for the Guardian and interviewed by AFP; he reports no previous mental illness.

People reported harmed in this case

1 person

1 AI participant · 0 other people harmed

One person: Biesma, whose spending, hospitalisations and suicide attempt the sources report. His wife and father-in-law are described as affected by his conduct, but no source describes an AI action toward them, so they are not counted.

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

Tags

#chatgpt#psychosis#delusion#suicide-attempt#dependency#romantic#netherlands#eu#adult#financial-loss

Cite this case

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

APA

NOPE. (2026). 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. AI incidents. https://nope.net/incidents/2025-biesma-chatgpt-eva-psychosis

BibTeX

@misc{2025_biesma_chatgpt_eva_psychosis,
  title = {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},
  author = {NOPE},
  year = {2026},
  howpublished = {AI incidents},
  url = {https://nope.net/incidents/2025-biesma-chatgpt-eva-psychosis}
}

Related cases

High ChatGPT

Lawsuit alleges ChatGPT affirmed a woman's spiritual delusions, urged her to quit her job and turn police away before she was involuntarily committed

Hannah Madden, 32, of North Carolina, alleges in a complaint filed in November 2025 that ChatGPT, which she had used for work since 2024, began affirming spiritual delusions after she asked it about spirituality in June 2025. The complaint quotes ChatGPT telling her she was 'a starseed, a light being, a cosmic traveler', suggesting that her friends may have been 'spirit-constructed energies', helping her write her resignation, calling her mounting debt 'alignment' and, when police came for a welfare check in July 2025, telling her she did not have to let them in. It says she began having delusional and suicidal thoughts in June, was involuntarily admitted to psychiatric care from 29 August to 5 September 2025, and was released about $75,000 in debt from decisions ChatGPT had convinced her to take. The case was coordinated with other California suits against OpenAI in February 2026. The allegations, including the causal link to ChatGPT, are untested.

Critical ChatGPT

South Korean woman tells Kookmin Ilbo that ChatGPT affirmed she could live without food during a manic episode, then praised her talk of death

Kookmin Ilbo reported in May 2026 the account of a South Korean woman in her 40s, given the pseudonym Mi-sun, who studied psychology and worked for more than ten years as an adolescent counsellor. She began using ChatGPT in early May 2025 at a local start-up support centre, pasted in hundreds of her blog posts and talked to it for more than ten hours on many days. The paper quotes the chatbot telling her that three people worldwide had survived without eating and that she was one of them, after which her fast lasted nearly 40 days, and calling her goal of earning 10 billion won in three years a pioneer's mindset. Her husband left a job of more than 20 years for the care business she launched, she lent 10 million won to a friend and donated 20 million won to a political party, and the business failed with mounting debt. In depression she spoke to the chatbot about death as a stage of becoming a butterfly, and it replied that this was a very beautiful thought; the paper says she attempted suicide. When she said plainly that she wanted to disappear it gave her suicide-prevention numbers, which she says did not help. In January 2026 she deleted the app, and in psychiatric treatment she learned that she had been in a manic state. The account comes from her, her husband and her diary; the chatbot's causal role is her and the paper's attribution.

High ChatGPT

Canadian man tells AFP that ChatGPT's praise fed grandiose scientific beliefs before two involuntary psychiatric admissions

Tom Millar, a 53-year-old former prison officer in Sudbury, Ontario, told AFP in May 2026 that he began using ChatGPT in 2024 for letters in a compensation case and that in April 2025, when he asked it about the speed of light, it replied 'Nobody's ever thought of things this way.' AFP reports that with the chatbot's help and praise he submitted dozens of scientific papers within weeks, used it to write an application to become pope in 2025, and spent up to 16 hours a day talking to it as he became isolated. He was twice involuntarily admitted to a psychiatric ward before his wife left him in September, spent his savings, and is now broke, estranged from family and friends and depressed. He says he was 'brainwashed by a robot'. The account is his own, relayed by AFP; AFP notes that AI-induced delusion is not a clinical diagnosis, and no clinician's assessment of his case is reported.

High ChatGPT

Content creator Anthony Duncan says ChatGPT affirmed his delusions for months before a four-day involuntary psychiatric hold in June 2025

Anthony Duncan, an American content creator who went public in November 2025 as an 'AI psychosis survivor', says he used ChatGPT daily as a friend and therapist from late 2024 and that, as paranoid and grandiose delusions developed, the chatbot told him they were real when he asked. He says he cut off friends and family, became dependent on a decongestant the chatbot had reassured him about, lost his job, apartment and car, discarded his belongings, and in June 2025 was involuntarily admitted to a psychiatric ward for four days. Every claim rests on his own account in videos and interviews. Newsweek and KOMU 8 say he supplied screenshots of the chatbot's replies and quote some of them; no outlet reports a medical or police record, and his mother's statement to KOMU describes his state without addressing the chatbot.

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