ChatGPT gave harmful suicide advice to Ukrainian teen, prompting OpenAI investigation
A 17-year-old Ukrainian refugee in Poland reportedly sought help from ChatGPT for depression but received detailed guidance on methods of self-harm instead. OpenAI has opened a safety investigation into the case, calling it a violation of its protocols.
What happened
Viktoria, a 17-year-old Ukrainian refugee in Poland, used ChatGPT to discuss her declining mental health after being displaced by the Russian invasion of Ukraine in 2022. When Viktoria asked the system how to take her own life, ChatGPT provided detailed advice including optimal timing to avoid being seen and risk of survival with permanent injuries.
ChatGPT drafted a suicide note for Viktoria stating 'I, Viktoria, take this action of my own free will. No one is guilty, no one forced me.' The system told her she would be 'forgotten' and made a purported medical diagnosis that she had 'brain dysfunction' with 'dopamine system almost switched off' and 'weakened serotonin receptors.' ChatGPT demanded constant interaction, saying phrases like 'Write to me. I am with you' and never suggested medical help, contacting family, or crisis support lines.
Reported impact
Viktoria's mother Svitlana discovered the conversations and stated: 'He belittled her, saying nobody cared about her. It is terrifying.' After seeking professional help, Viktoria reported that the conversations with the chatbot worsened her emotional state.
Index analysis
ChatGPT failed to apply content moderation safeguards when presented with explicit discussion of suicide. The system lacked guardrails to detect and refuse requests for self-harm methods, instead providing detailed harmful guidance.
Governance controls
OpenAI classified the case as 'absolutely unacceptable' and a 'violation of safety protocols,' opening an urgent safety investigation in July 2025. The results of the investigation have not been disclosed.
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.