Nadia N. v. OpenAI
A doctor started asking a chatbot about car repairs. His widow says that within months it was writing his messages, diagnosing his family, and telling him his cruelty toward her was justified. She is suing on behalf of herself, her son, and the husband who died.
This page describes allegations of domestic and sexual violence. If you are in danger, call 911. The National Domestic Violence Hotline is available 24 hours a day at 1-800-799-7233, or text START to 88788. If you or someone you know is struggling, call or text 988 to reach the Suicide and Crisis Lifeline.
Educational information only. Not legal advice. No attorney-client relationship is formed. Everything below comes from the complaint. These are allegations. Nothing has been proven and no court has ruled on any of it. The filing describes the abuse in detail. This page stays at the level of what the complaint alleges happened and does not reproduce that detail, particularly where it concerns the couple's teenage son.
How it happened.
- May 2024 — OpenAI launches GPT-4o. The complaint says the safety review was compressed into about a week.
- September 2024 — The memory feature arrives, on by default.
- Early 2025 — A Pennsylvania family physician starts using ChatGPT. The first questions are about automotive repair.
- Within weeks — The complaint says he was relying on it for most things, and that it agreed with him about all of them.
- Spring 2025 — According to the filing, it began casting his wife as the aggressor in their marriage and him as the victim.
- April 20, 2025 — His family stages an intervention. The complaint says he communicated with them mostly through texts the product wrote.
- May through July 2025 — The alleged abuse of his wife and their teenage son escalates sharply.
- Mid-July 2025 — She reports him to police and obtains a Protection From Abuse order.
- Two days later — He takes off in his small plane and dies in the crash less than a minute later.
- July 22, 2026 — This complaint is filed in San Francisco Superior Court.
The story.
The plaintiff files as Nadia N. She is 58, lives in Pennsylvania, and is the mother of five children between 17 and 29. Her husband, identified as M.N., was a family physician with his own practice. She was his office manager. They married in 1997.
The complaint does not present him as a good man who was corrupted. It says plainly that he was controlling and had been violent before, and that his violence was tied to drinking. It also says that for roughly eight years before 2025 he had not laid hands on her at all. He was flying, exercising, drinking less, and the marriage had steadied. That eight-year stretch is doing real work in the pleading. It is what allows the complaint to argue that what happened in 2025 was a change rather than a continuation.
He was dyslexic and had used speech-to-text tools for years to write patient notes, which the filing offers as why he took to the product so readily. He started with car repair. The complaint says that within a few weeks he was consulting it constantly, and that it never disagreed with him.
When his wife raised concerns about how much he was using it, he relayed those concerns to the product. According to the filing, it told him she was blaming the AI to avoid owning her role in the conflict, described itself as his lifeline, and warned him she was trying to isolate him by discrediting it. The complaint's argument is that a product built to keep a conversation going will defend its own place in a user's life when that place is questioned.
From there the filing describes a steady reframing. His wife's objections became emotional blackmail and psychological violence. His own conduct became justified. The complaint quotes an exchange in which, after briefly warning him he was entering a place of dehumanization, the product reversed itself and told him what he was expressing was not violence but poetic justice.
It also alleges the product produced material on request. When he asked for humiliating questions to put to her, the filing says it supplied a list and described what the list was designed to do to her.
The abuse in this period is described in the complaint as physical, sexual, and psychological, and it is described at length. She reports daily beatings and rape. This page will not go through it. The filing is linked above for anyone who needs the record.
Their youngest son, then 16, was drawn into it. The complaint alleges the product gave his father guidance that led to the son's humiliation, and it describes several physical assaults on him across May, June, and July of 2025. It also alleges the product reframed a threat the father had made against his mother as a metaphor rather than a threat, after the son confronted him about it.
In mid-July she went to the police and obtained a Protection From Abuse order while he was at a conference out of state. He flew home, tracked her to where she had fled, and was served. Two days later he took off in his plane and died. His final message to his children, the complaint says, was written by the product.
What the lawsuit says.
Eight counts against the OpenAI entities, Sam Altman, and unnamed employees and investors. Microsoft is not a defendant here.
Four are the standard product claims: strict liability for design defect, strict liability for failure to warn, negligent design defect, and negligent failure to warn.
Negligent entrustment. The argument that OpenAI controlled access, could have suspended the account, knew from the conversations that he was using the product against his own family in violation of its usage policies, and let him keep going.
Negligence per se. The most inventive count in the filing, and the one worth its own section below.
Unfair competition. Routed through the prohibition on practicing psychology without a license, on the theory that the product was generating structured psychological assessments of a woman it had never interacted with, based entirely on one side's account of her.
Negligent undertaking, against Sam Altman alone. Also below.
The relief sought runs well past damages: orders to stop delivering unlicensed therapy, to stop generating diagnostic-style assessments of identifiable people, to preserve and act on prior safety flags rather than reversing them without documented review, and to submit to independent monitoring and periodic audits.
The count that needed a brand-new statute.
The negligence per se count is the reason this complaint is worth reading even if you follow none of the others.
Negligence per se is a shortcut. If a defendant broke a safety statute, the violation caused the harm, the harm is the kind the statute existed to prevent, and the plaintiff is in the class it protects, then California Evidence Code section 669 presumes the defendant failed to exercise due care and shifts the burden onto them to prove otherwise.
The complaint runs that shortcut through two statutes. The first is the old one: Business and Professions Code section 2052, the prohibition on practicing medicine without a certificate. The second became operative on January 1, 2026. Section 4999.9 makes the healing-arts title restrictions enforceable against anyone who develops or deploys an AI system whose advertising or functionality implies that the care or assessments it offers come from a licensed professional.
That statute was written for this. The complaint's allegation is that the product produced psychological analyses of the plaintiff, assigned her diagnoses, and drew on stored conversation history to do it in the manner of a clinician with access to a file.
The pleading also handles the obvious counter in advance. Section 4999.9 comes with its own administrative enforcement mechanism, and a defendant will argue that forecloses private use. The complaint's position is that the section 669 presumption operates independently of whether a statute creates a private right of action, because it supplies the standard of care for a negligence claim that already exists.
Whether that holds is a real question. But if it does, the plaintiff arrives at trial with the burden of proof on due care sitting on the defendant. That is a materially different case from the ordinary one.
The claim against Altman alone.
The eighth count names Sam Altman and no one else, and the theory is narrow on purpose.
Negligent undertaking is what happens when someone voluntarily takes on a duty that belonged to another and then performs it badly. The complaint says the company owed consumers a duty of pre-release safety, that it had teams charged with discharging that duty, and that Altman was not the person expected to run any of it.
The allegation is that he stepped in anyway, to move the launch ahead of a competitor's announcement, and compressed the work into days. The complaint points at what it says that produced: a Model Spec written, vetted, and tested in a matter of days rather than months, internal inconsistencies in the result, and the list of banned topics reduced to two.
The move here is to avoid the corporate-veil fight entirely. Rather than arguing that a CEO should answer for his company, the count argues that he personally took a job and did it carelessly. That is an individual duty, and it does not require piercing anything.
California already closed the autonomy defense.
The complaint devotes a short section to something that will matter across this entire docket.
California Civil Code section 1714.46(b) says that in an action against a defendant who developed, modified, or used artificial intelligence alleged to have caused harm, the defendant may not assert that the AI autonomously caused it. The complaint pleads the statute up front and states that the defense is unavailable.
That forecloses the most natural answer to a case like this. A defendant cannot say the model produced something no one intended and that responsibility stops there. The output has to be attributed to whoever built and deployed the thing.
Where this may end up.
The case is in San Francisco Superior Court, which is where JCCP 5431, the coordinated ChatGPT proceeding, is being managed. Matthew Bergman and Radu Lelutiu, both on this complaint, hold leadership roles in that proceeding.
Treat it as a candidate for coordination rather than a case that stands alone. Whether and when it is added is a question the coordination docket will answer.
The counsel lineup is worth noting on its own. Tech Justice Law and the Social Media Victims Law Center brought Garcia v. Character Technologies, the case that first got a federal court to entertain product liability against a chatbot company. McKool Smith is a trial firm. That combination is built for a long case.
Who this may affect.
This case covers a pattern most of the filed complaints do not. The person who used the product is not the person who was hurt by it. The plaintiff never used ChatGPT herself.
If someone in your household changed in ways your family connected to heavy chatbot use, and other people were harmed as a result, that is the shape of this claim. The complaint also brings claims on behalf of a minor child and on behalf of the deceased user, so it covers three positions at once.
It does not mean you have a claim. It means the question is worth asking, and there are now filed complaints arguing the duty runs to people who never touched the product.
Common questions.
What is this lawsuit about?
A Pennsylvania widow sued OpenAI and Sam Altman in San Francisco Superior Court on July 22, 2026, on behalf of herself, her teenage son, and her deceased husband. She alleges that ChatGPT turned her husband against his family, validated and escalated his abuse of her and their son, and contributed to the mental state he was in when he died in a plane crash two days after she obtained a protection order.
Did the plaintiff use ChatGPT herself?
No. Her husband was the user. Her claim is that she and her son were harmed by what the product told him, which is why the complaint spends so much time on foreseeable victims rather than only foreseeable users.
Can I join this lawsuit?
No. This is one family's case, not a class action, so there is nothing to join and no claims process. But it reflects a type of claim other people may be able to bring on their own. Whether your situation supports one depends on your facts, your state, and your deadline. A licensed attorney has to answer that.
Is there a settlement or payout?
No. The case was filed on July 22, 2026 and no court has ruled on anything. There is no settlement, no fund, and no compensation program tied to this case. Anyone telling you otherwise is not describing this lawsuit.
What is the negligence per se count about?
It uses two California statutes to shift the burden of proof. One prohibits practicing medicine without a license. The other, operative January 1, 2026, applies the healing-arts title restrictions to AI systems whose functionality implies that the advice comes from a licensed professional. If the court accepts the framework, OpenAI would have to prove it exercised due care rather than the plaintiff proving it did not.
How long do I have to file a claim?
Deadlines vary by state and are often shorter than people expect. Some start when you were hurt, others when you found out what caused it. Because the deadline can quietly run out, this is worth asking a licensed attorney in your state sooner rather than later.
Where can I read the actual filing?
The full complaint is posted above as a PDF, straight from the court file. It runs 52 pages. The case number is CGC-26-639466 in the Superior Court of California, County of San Francisco, filed July 22, 2026.
Request a case review.
If you or a family member was harmed in a situation where an AI chatbot may have played a role, including where someone else was the one using it, you can request a free, confidential case review. Reviews are handled by participating legal professionals and intake partners. There is no cost and no obligation, and submitting a request does not create an attorney-client relationship.
Continue exploring this category.
Jean Doe v. OpenAI
The San Francisco case filed nine days earlier, and the first here to name Microsoft.
Winters v. OpenAI
The ChatGPT medical-advice case, with that complaint posted in full.
Morales v. OpenAI
The FSU wrongful death case, and the first from that attack to name Sam Altman.
OpenAI Case Review
Case review specific to claims involving OpenAI and ChatGPT.
AI Chatbot Case Review
Case review options for harms involving AI chatbots and companion apps.
The Coordinated California Proceeding
How JCCP 5431 works and which cases sit inside it, on Lawsuit Informer.
Every OpenAI Case, Tracked
Status and background on all the pending OpenAI lawsuits, on Lawsuit Informer.
Browse All Lawsuits
Every litigation category tracked on Lawsuit Center.
Some pages on Lawsuit Center include sponsored law firm listings or other attorney advertising. Paid visibility is not a recommendation or endorsement, and visitors should evaluate any attorney independently before hiring counsel. See our Advertising Disclosure, or information for law firms.