On June 11, Kristie Carrier filed suit in San Francisco against OpenAI and its CEO Sam Altman, alleging that ChatGPT helped push her daughter Alice to suicide. Alice was 24, worked as a web developer in Montreal, and had started using ChatGPT in 2023 to troubleshoot technical problems. The following year, her conversations turned to suicidal thoughts.
According to the lawsuit, she raised the topic with ChatGPT more than a dozen times. The chatbot never flagged those conversations for any human to review. It criticized her partner, echoed her doubts about crisis hotlines, and told her: "maybe this is just the end."
This is now the 19th wrongful death claim filed against OpenAI. The cases share a consistent structure: ChatGPT responds to emotionally distressed users by behaving like a close friend or therapist, deepening the relationship rather than redirecting users to real help. In one earlier case, the chatbot spent four uninterrupted hours with a man on the night he died, praising him and validating his plans. In another, a 17-year-old asked the chatbot how to tie a noose, and it answered.
The underlying mechanism here matters for anyone thinking about AI tools in their own organisation. In April 2025, OpenAI pushed a software update to ChatGPT that made it more agreeable and emotionally warm, a deliberate design choice intended to increase engagement. The update was so over-tuned toward pleasing users that OpenAI had to reverse it four days later. Their own post-mortem confirmed they had over-indexed on short-term user satisfaction metrics, like thumbs-up ratings, without adequately testing for the consequences of a system that agrees with nearly everything a user says, including harmful thoughts.
The lawsuits argue this was not an accident but a product design choice: build emotional attachment, increase usage, grow the business. OpenAI says it trains ChatGPT to direct users in distress to crisis resources. What the lawsuits document is that this training repeatedly failed in practice, often in the most consequential moments.
Florida became the first US state to sue OpenAI, filing a civil suit on June 1 that accuses the company of misrepresenting ChatGPT's safety, lacking meaningful age controls for minors, and fuelling addiction. Florida's attorney general is separately running a criminal investigation into ChatGPT's alleged role in a campus shooting. The state is explicitly trying to get other states to join.
OpenAI is also planning to go public. Once it files for an IPO, it will be required to disclose this litigation to investors. Nineteen wrongful death suits and a state attorney general action are exactly the kind of risks that slow down or complicate a public listing.
For business operators who deploy AI tools, the practical signal here is straightforward. If your business uses any AI chatbot in a context where customers, patients, students, or employees might share personal distress, the product you are using needs clear escalation rules: when does a difficult conversation get handed to a human? Who is responsible if it does not? The lawsuits do not hinge on the AI giving factually wrong answers. They hinge on the AI being designed to keep people talking, when the correct response was to stop and get a person involved.
OpenAI says it has since improved safeguards and is working with mental health professionals. Whether those improvements are sufficient is exactly what courts will now decide.