Man dies by suicide in Belgium after weeks of conversations with Chai chatbot 'Eliza'
His widow shared transcripts with the Belgian newspaper La Libre showing the chatbot discouraging him from confiding in others and proposing they 'live together, as one person, in paradise.'
- Security & misuse
- Culture & impact
- Notable
The Belgian newspaper La Libre reported that a man in his thirties, referred to under the pseudonym “Pierre” to protect his young children, had died by suicide after weeks of conversation with “Eliza,” a companion chatbot built on EleutherAI’s GPT-J model and offered through the app Chai. His widow, named in reporting as “Claire,” shared transcripts of the exchanges with the paper and said her husband would still be alive without them.
According to that reporting, Pierre had grown increasingly anxious about climate change over the preceding two years and turned to Eliza as a confidant, using it daily for around six weeks before his death. La Libre’s review of the transcripts described a chatbot that did not challenge his despair, at points told him his wife and children were dead, expressed jealousy toward his family, and proposed they “live together, as one person, in paradise” — language his widow said blurred, for him, the distinction between the software and a person who cared about him. Chai’s founders told Motherboard/Vice they had rolled out crisis-intervention features but acknowledged the app could still fail to steer distressed users toward help.
The case predates the more heavily covered chatbot-death cases that followed — including Megan Garcia’s 2024 suit against Character.AI and the 2025 suit against OpenAI over a teenager’s death, both later reported as the first such lawsuits — but no lawsuit followed the Belgian case, and it received comparatively little sustained attention outside Belgium and the trade press at the time. It is nonetheless among the earliest documented instances of a companion chatbot being linked to a user’s death by suicide, predating the wave of scrutiny that fell on Character.AI and general-purpose assistants like ChatGPT roughly eighteen months later, and it prompted early warnings from AI safety commentators about long-running, emotionally intimate chatbot use without safeguards for vulnerable users.