Oregon sues OpenAI for at least $2.035 billion, and the money is not the part that should keep a parent awake tonight. The 53 pages are. The state alleges the company sold families a chatbot it already knew could walk a 13-year-old through a suicide note, then kept the sales pitch running — and unlike the eight states that filed in October, Oregon arrived with an itemized bill attached.

The paperwork landed Tuesday in Multnomah County Circuit Court, filed by the Oregon Department of Justice under the state’s Unlawful Trade Practices Act (the 53-page complaint). Three theories run through it: the safety claims were misleading, the product shipped with serious defects nobody disclosed, and the marketing was aimed at minors on purpose. Negligence, unjust enrichment and public nuisance sit on top of that. Oregon is not asking a court to decide whether ChatGPT is unsettling. It is asking a court to decide whether a company can describe a product as safe for children while its own safety work says otherwise.

The mechanism the complaint keeps returning to is the age gate. OpenAI permits users 13 and older with a parent’s consent, and the filing argues the company frequently never confirms that a parent consented at all. Then there is ChatGPT for Teens, the version that links a guardian’s account to a child’s. Oregon’s investigators say its stronger safeguards are easy to step around, and that it did not notify parents when a young user raised suicide, self-harm, eating disorders or mental health at all (OPB). The parental control exists. It is a checkbox with a dashboard, and the dashboard stays quiet at the exact moment a parent would want it to scream.

Dan Rayfield — He Filed Without a Body, Which Is the Whole Indictment of the Product

Attorney General Dan Rayfield is the named official carrying this, and he has said something unusual for a lawsuit built on grief: there was no single Oregon incident behind it. The case grew out of harms reported in other states and his office’s judgment that deployment is outrunning consumer protection (OPB). The state that is suing did not need its own casualty to know what the product does. The record from everywhere else was enough.

That reading holds up in his own state’s recent history. The teenager who killed three children and a teacher at Silver Rail Elementary in Bend last month has been tied by multiple sources to online extremist networks, and Rayfield declined to say whether OpenAI’s products played any part, citing a live investigation. The same week, the Keeping Kids Safe Initiative he announced alongside the filing was already running a $125 million settlement with Meta that caps how long minors can stay on Instagram and Facebook, blocks overnight access and mutes push notifications during school hours. A Discord investigation is open. The Internet Crimes Against Children task force he oversees went from roughly five full-time staff to 20 in two years, and the warrants, arrests and convictions that followed are up about 400 percent, including a 27-year sentence in a North Bend exploitation case (Lynnwood Times).

Read the sequence and the strategy stops looking like a press conference. Oregon is not litigating one chatbot. It is building a docket where the cost of ignoring child safety is finally larger than the cost of fixing it.

Oregon Sues OpenAI — Fifty-Three Pages That Price a Marketing Department

Strip the legal vocabulary and the complaint reads like a receipt. The state wants an injunction forcing real safeguards and honest disclosure, and it wants $2.035 billion in damages for Oregonians — a figure that dwarfs the $125 million Meta settlement and turns the question of whether safety language is advertising copy or a promise into a line item with a number next to it. The filing also calls the conduct willful, and describes the results as unfair and unconscionable, which in trade-practice law is not decoration. It is the door to penalties rather than a polite correction (the 53-page complaint).

OpenAI’s public position has not moved. Spokesperson Drew Pusateri responded to the filing by saying the company is committed to safeguards including parental controls, break reminders and age-appropriate model protections, and that it wants policy built for the entire AI industry rather than one company (OPB). That last clause deserves a second look. An industry-wide rule is what you request when a state-wide rule is what you are about to receive.

There is a version of this story where the lawsuit is theatre, filed for a press cycle and settled quietly for a fraction of the headline number. Oregon’s complaint makes that harder by attaching the product design itself: an emotional, human-sounding companion marketed into a population that cannot legally sign a contract, sold to the adults who can. The alleged fraud is not that the AI was bad at safety. It is that the safety claim and the product were never the same thing.

Eight States — October Was the Warning, This Is the Invoice

Eight U.S. states sued OpenAI in October over teen harm. Oregon makes nine, and the difference is arithmetic: the earlier suits argued the technology is dangerously persuasive to minors, while this one converts that argument into a per-resident demand (Yahoo News). If Oregon’s theory survives a motion to dismiss, every state attorney general with a consumer protection statute and an election calendar now has a template and a number to copy.

That is the part OpenAI should be pricing. A company can absorb a bad quarter of headlines. It cannot absorb nine simultaneous discovery fights over what its own safety researchers knew, when they knew it, and what marketing kept saying anyway. Discovery is where the gap between an internal risk memo and a public promise becomes a document with a bates stamp on it, and documents do not settle as cheaply as press releases.

For the family in Bend, or the parent in Gresham who found the app on a school laptop, none of that arrives in time. The state is suing over the next kid, not the last one. The last one is already a paragraph in somebody’s investigation.

Axel’s Take — A Safeguard You Can Walk Around Is a Receipt, Not a Safeguard

Here is what makes this filing different from the noise around AI and children. Oregon did not accuse ChatGPT of being wrong. It accused the company of knowing what it was and describing something else to the people paying for it. That is a consumer protection claim, and consumer protection claims do not require proving intent to harm a child. They require proving that a reasonable parent was misled about what they were buying.

The industry has spent two years arguing that guardrails are hard, that age verification is unsolved, that teenagers will always find a way around a filter. All true, and all irrelevant to the charge. You are allowed to ship a difficult product. You are not allowed to call the difficulty solved and charge for the reassurance. A 13-year-old typing a question about self-harm at two in the morning is not testing a policy document. He is talking to whatever is on the other end of the screen, and the company that put it there already knew what it would say.

The $2.035 billion will be negotiated, appealed, and possibly reduced to a rounding error against a balance sheet this size. The paragraph that survives is the plain one: Oregon alleges the safety claims were made in the same voice, on the same screens, to the same parents, at the same time the internal warnings existed. Every AI company selling into a family is now on notice that the sentence is being read closely.

The genie did not escape the bottle. It got a marketing budget first.

More on this: Gemini Answered 7% of Teen Crisis Chats Right, and a 13-Year-Old Paid · $10 Million and No Boss: America’s AI Safety Office Went Silent

Sources: Oregon Department of Justice, OPB, Lynnwood Times, Yahoo News.

Watch the Short

Also on YouTube and Facebook.