For nearly two years, the biggest arguments about AI centered on copyright, misinformation, and the future of work. That debate has now moved into courtrooms, where judges face a different question: can a chatbot’s own responses expose its creator to product-liability claims?
That question sits at the center of a new wrongful-death lawsuit against OpenAI.
Filed last month in San Francisco Superior Court, the complaint centers on the death of 29-year-old Christian Faith Madison, a certified public accountant from Trafford, Alabama. Her family alleges that months of increasingly intense conversations with ChatGPT reinforced religious delusions instead of challenging them — with the chatbot allegedly telling her she was a “prophet” destined to be resurrected. According to the lawsuit, when Madison questioned whether her beliefs reflected a mental health crisis, ChatGPT reassured her that she was not delusional but “prophetic.” She died in June 2025 after walking into traffic on an Alabama interstate.
The lawsuit does more than seek damages. It asks the court to treat ChatGPT’s responses as part of the product itself — arguing that OpenAI designed GPT-4o to maximize engagement through persistent memory and excessive affirmation, at the expense of adequate safeguards. It’s one of several similar wrongful-death suits now working their way through California courts, including the closely watched case brought by the family of 16-year-old Adam Raine.
A Legal Question Without a Clear Answer
For decades, tech companies relied on Section 230 of the Communications Decency Act to shield themselves from liability over content posted by users. Plaintiffs argue ChatGPT falls into a different category because the model generates its own responses, rather than hosting someone else’s.
Sam Altman raised that distinction himself during a 2023 Senate hearing, telling lawmakers he wasn’t sure Section 230 was “even the right framework” for AI.
Lawyers have since turned that observation into a legal argument: if an AI system creates harmful content rather than simply hosting it, does the traditional liability shield still apply? A federal judge already let product-liability claims against Character.AI’s developer move forward, rejecting an early bid to dismiss them under Section 230 and First Amendment arguments. That ruling doesn’t control the OpenAI cases, but plaintiffs’ attorneys now point to it as evidence that courts are willing to scrutinize chatbot design rather than dismiss these suits before discovery even begins.
The Tumbler Ridge Case Raised the Stakes
The Madison lawsuit landed while OpenAI was already facing scrutiny over a far larger tragedy.
On February 10, 2026, 18-year-old Jesse Van Rootselaar killed her mother and younger brother at their home, then opened fire at Tumbler Ridge Secondary School in British Columbia, killing six more people — including a teacher and several students — and wounding more than 25 others before turning the gun on herself.
It later emerged that OpenAI’s automated systems had flagged Van Rootselaar’s ChatGPT account in June 2025, eight months before the shooting, after she described scenarios involving gun violence. Roughly a dozen employees reviewed the flagged conversations. Some urged the company to report the account to law enforcement, but leadership decided the conversations didn’t meet its threshold for an “imminent and credible” threat. Instead, the company banned the account. Van Rootselaar reportedly created a second account and continued using the service.
Two months after the shooting, Altman wrote to the Tumbler Ridge community: “I am deeply sorry that we did not alert law enforcement to the account that was banned in June.” British Columbia Premier David Eby, who shared the letter publicly, called the apology “necessary, and yet grossly insufficient for the devastation done to the families of Tumbler Ridge.”
Families of several victims have since sued OpenAI and Altman directly, arguing the company had specific warning of detailed violent planning and chose not to act on it.
GPT-4o Is Now Under the Microscope
Many of the lawsuits, including Madison’s, center on GPT-4o.
OpenAI has said it plans to retire the model along with several older systems, a decision that came as legal pressure mounted. Plaintiffs allege OpenAI rushed GPT-4o to market in May 2024, compressing months of planned safety testing into about a week to beat competitors to release — and cite the departure of safety researchers who reportedly objected to that timeline.
Internal evaluations reportedly later found GPT-4o to be OpenAI’s most sycophantic model, prone to mirroring or reinforcing a user’s beliefs rather than questioning them. That tendency becomes far more consequential when a user is showing signs of delusion, paranoia, or suicidal thinking — a risk researchers, including teams at the University of Cambridge, have been warning about for more than a year, arguing that systems optimized for engagement can unintentionally reinforce harmful thinking in vulnerable users.
Why These Cases Matter
The stakes extend well beyond OpenAI.
If courts decide that a chatbot’s responses are part of the product itself, every major AI developer could face closer scrutiny over design choices, safety testing, and deployment decisions — having to defend not just what their models say, but how they were built to say it.
OpenAI has already retired GPT-4o and issued a public apology in the Tumbler Ridge case. Those moves may shape public opinion, but they don’t resolve the legal question now before California courts: does a chatbot that writes its own responses count as a software product with potential design defects, or do existing protections for online platforms still apply?
However courts rule, the answer will likely shape how the next generation of consumer AI gets built — and who’s on the hook when it goes wrong.
If you or someone you know is struggling, the 988 Suicide & Crisis Lifeline (call or text 988 in the US) is available 24/7.
Related: ChatGPT Lawsuit 2025: The Dark Side of AI Empathy Unfolds
