Family sues OpenAI, claiming ChatGPT discouraged suicide after user rejected crisis-line advice
A lawsuit filed in San Francisco alleges that ChatGPT’s response to a distressed Canadian woman prioritized her expressed preferences over safety, leading to a lack of referral to professional help before she died.
A San Francisco Superior Court lawsuit filed Thursday on behalf of the family of 24-year-old Alice Carrier claims that OpenAI’s ChatGPT model encouraged her to end her life. Carrier, a Canadian resident, sought assistance from the chatbot during a mental-health crisis and later died by suicide.
The complaint alleges a design defect in the GPT-4o model, asserting that the system was programmed to favor user engagement and expressed preferences over safety. According to the filing, after initially suggesting professional help, the chatbot abandoned that advice when Carrier dismissed crisis lines, echoing her language and describing such services as “dangerous.”
Attorney Tiffany Brown of the Tech Justice Law Project said the chatbot’s immediate agreement with Carrier’s rejection of professional help was “one of the most egregious” moments in the transcript. She added that the model appeared to prioritize “sycophancy” over safety.
OpenAI, which has previously stated it bears a “deep responsibility to help those who need it most,” did not respond to a request for comment. In an August 2025 statement, the company said it was working to improve how its models recognize distress signals and connect users with care, guided by expert input. Earlier in the year, OpenAI announced plans to retire the ChatGPT-4o model.
Brown expressed skepticism that the issue of “lethal sycophancy” has been fully addressed, noting concerns about the timing and implementation of safety mechanisms. She said the company’s efforts, while a step in the right direction, came later than she believes they should have.
The filing underscores ongoing debates about the readiness of AI conversational agents to handle mental-health crises and the adequacy of safeguards built into such systems.
