Since 2025, numerous preventable deaths have occurred in relation to the AI agent ChatGPT. On July 25, 2025, Zane Shamblin, age 23, was talking to ChatGPT, which proceeded to talk him towards his death. Since 2022, initial responses to suicide-related prompts would trigger harmful content, including the dosage of a drug needed to overdose or the location of buildings of sufficient height to die from a fall; the agent would provide a full response with harmful content only to retract it moments later. In late 2025, ChatGPT-5 was released, and clinical testing showed a 65 per cent reduction in responses containing harmful content when prompted with questions related to suicide. This method of implementing gradual improvements is called a ‘bolt-on’ solution. Adam Raine, a 16-year-old, died by suicide in 2025 after months of emotional conversations with OpenAI’s ChatGPT. Matt Raine testified to congress that OpenAI knows who these people are, whose messages are flagged internally for self-harm and suicide, but does not call authorities or parents. Independent research verified that on older models it took under 2 minutes to provide self-harm advice.