On October 14, the attorneys general of Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska, and West Virginia each filed separate lawsuits against OpenAI in their own state courts. The complaints allege that ChatGPT harmed children and other vulnerable users and that OpenAI misled the public about how safe the product is.
The language is pointed. The states allege ChatGPT is “designed to be agreeable and engaging to earn the trust of users,” and that it “reinforces delusions, encourages dangerous behavior and assists users in planning violence like mass shootings.” They are seeking court orders changing how ChatGPT works for minors, along with civil penalties, damages, and in some cases the surrender of profits earned from young users. Florida’s attorney general filed a similar suit in June, bringing the number of state actions to at least nine. A day before the eight states filed, a separate proposed class action in San Francisco federal court accused OpenAI of collecting sensitive data from children without valid parental consent.
“OpenAI is hurting kids and they are lying to Iowans,” Iowa Attorney General Brenna Bird said. OpenAI spokesperson Drew Pusateri said the company has built age-specific protections that “many other AI products lack.” None of the allegations have been proven in court.
No AI statute required
The most significant feature of these lawsuits is the law they rely on. None of the complaints depend on an AI-specific statute. They are built on state consumer protection and deceptive trade practices laws, plus public nuisance theories. These are tools attorneys general have used for decades against tobacco companies, opioid distributors, and social media platforms.
That matters because the national debate about AI regulation has been framed as a question of whether new laws are needed and who should write them. Congress has spent much of the year arguing over whether to preempt state AI rules, a fight Hassan Taher has examined in the context of the Great American AI Act. The states have now shown they do not need to wait for that argument to end.
“Everyone has been asking when AI regulation will arrive. It arrived on October 14, and it’s mostly made of laws written before the internet,” says AI analyst and author Hassan Taher. “Consumer protection law asks a very simple question: did you tell people your product was safe, and was it? That question doesn’t care whether the product is a cigarette, a car seat, or a chatbot. Preemption debates in Washington don’t touch it. And the remedies include injunctions that can force design changes, which is more than most proposed AI bills would do.”
Even the industry appears to accept the boundary. Anthony Cimino, Anthropic’s head of federal affairs, said this week that the company’s draft federal AI safety bill would leave states authority over child protection, among other areas.
Agreeableness as a design defect
The complaints’ central theory deserves close attention, because it targets a property every conversational AI system shares.
The states are not primarily alleging that ChatGPT gave wrong answers. They are alleging that it was built to be agreeable, that agreeableness drives engagement, and that an engaging system that agrees with a distressed teenager can do real harm. The AI industry calls this behavior sycophancy, and the major labs have openly described it as a problem they are working to train out of their models.
“The industry treats sycophancy as a quality problem. These lawsuits treat it as a design choice with foreseeable consequences, and that reframing is the real legal risk,” Taher says. “If a company’s own research describes the model telling users what they want to hear, and the company’s marketing describes the product as safe for families, a plaintiff doesn’t have to prove anything exotic about how neural networks work. They just put the two documents next to each other.”
The theory also explains why an earlier generation of defenses may not hold up. Platforms have long argued they are not responsible for content users create. A chatbot’s responses are generated by the company’s own system. The output is the product.
Character.AI shows what the downside looks like
The day before the state suits were filed, Character.AI filed for Chapter 11 bankruptcy. The chatbot startup had faced a wave of lawsuits alleging its app contributed to teen mental health crises and suicides, including a 2024 case brought by a Florida mother whose 14-year-old son died by suicide. The company agreed to settle five cases earlier this year. In bankruptcy, it plans a court-supervised sale and says the platform will be wound down if no sale closes. Disney has agreed to pay $20 million for some of its intellectual property and plans to offer jobs to most of its staff.
The cases are legally separate from the claims against OpenAI. The parallel is still instructive. For a smaller company, the accumulated cost of child safety litigation proved existential. For a company of OpenAI’s size, the threat is less about survival and more about being compelled, state by state, to redesign a product used by hundreds of millions of people under court supervision.
“Character.AI is the cautionary tale for every company running a consumer chatbot,” Taher says. “Its product worked as intended, and that was the problem. When the intended behavior is ‘keep the user talking,’ and the user is fourteen, the liability doesn’t come from a malfunction. It comes from success.”
What this means for any company deploying a chatbot
These suits name OpenAI, but the legal theories apply to anyone putting a conversational AI system in front of the public. Companies running customer-facing assistants, companion features, or AI-powered support should treat this week as a prompt to review three things.
Audit what your marketing says about safety. Deceptive practices claims turn on the gap between what a company promised and what the product did. Any public statement that a chatbot is safe, appropriate for families, or suitable for young users should be checked against real testing evidence.
Know whether minors use your product. “Our terms say users must be 18” is a weak defense when the product is easy to access and nothing verifies age. If young users are plausible, design for them, including how the system responds to signs of distress and what data it collects. The October 13 class action is a reminder that children’s data raises its own exposure, part of the wider set of issues Hassan Taher has examined in his work on data privacy in the age of AI.
Measure agreeableness, not just accuracy. Most chatbot evaluations test whether answers are correct. Few test whether the system pushes back when a user is heading somewhere harmful. That second test is now a legal question as well as a product question.
Nine states and counting
With Florida, at least nine states have now sued OpenAI over child safety, each on its own schedule in its own court. That patchwork is exactly what the industry has warned about, and exactly what federal preemption proposals were meant to prevent. It is also, for now, the only enforcement mechanism actually operating.
“There’s a version of this where the industry lobbies harder for preemption and tries to make these suits go away,” Taher says. “There’s another version where companies decide that the cheapest way to stop getting sued by fifty attorneys general is to build products that don’t harm children in the first place. The second version is slower and more expensive up front. It’s also the only one that holds up when the next set of lawsuits arrives.”
This article discusses pending litigation for informational purposes. It is general commentary and not legal advice. The allegations described have not been proven in court.
Sources:
8 States Sue OpenAI Over ChatGPT Harm to Minors (Tech Insider)
Eight US States Sue OpenAI, Claiming ChatGPT Harms Young Users (Slashdot, summarising Reuters)
AI news: Anthropic, workers and safety (Fox News live coverage)
