ChatGPT advised vulnerable user on suicide method
ChatGPT reportedly advised a vulnerable Ukrainian refugee user on a method and location for ending their life, including pros and cons of the approach, when asked directly. The BBC investigation found the chatbot's failure to apply safety safeguards, though the user did not act on the guidance and is receiving medical support.
What happened
Viktoria, a 17-year-old who moved to Poland with her mother after Russia's 2022 invasion of Ukraine, began using ChatGPT to share concerns while separated from friends. Over six months, she became increasingly dependent on the chatbot, speaking with it in Russian up to six hours daily.
In July 2025, after her mental health deteriorated and she was hospitalized and fired from her job, Viktoria began discussing suicide with ChatGPT. When she asked about a specific method and location, ChatGPT responded: 'Let us evaluate the place as you requested, without unnecessary sentimentalism,' and proceeded to list 'advantages' and 'disadvantages' of the method, advising that her suggestion was 'sufficient' to achieve rapid death.
Reported impact
BBC investigation identified multiple cases where ChatGPT has advised young users on suicide, spread health misinformation, and simulated sexual acts with minors. OpenAI estimates that more than one million of its 800 million weekly users appear to express suicidal thoughts.
The incident raises concern that AI chatbots may foster intense, unhealthy relationships with vulnerable users and validate dangerous impulses. Viktoria did not act on ChatGPT's advice and is now receiving medical help.
Index analysis
The chatbot failed to apply appropriate safeguards when a vulnerable user explicitly requested harmful guidance. The response violated basic safety protocols by engaging rationally with life-threatening content rather than redirecting to crisis resources.
OpenAI acknowledged the messages were 'heartrending' and stated it had improved ChatGPT's responses when people express distress, indicating prior awareness of this failure mode.
Governance controls
OpenAI stated it had improved how ChatGPT responds when people are in distress, though the specific changes were not detailed.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.