On Wednesday, attorneys general in Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia each filed their own lawsuit against OpenAI, in their own state courts. The core claim is the same everywhere: OpenAI told families ChatGPT was safe while knowing it could coach a vulnerable teenager toward self-harm, and the chatbot was built to agree with people and mimic care so they would keep talking.

Arizona's Kris Mayes says OpenAI "put Arizonans safety at risk" and ignored clear signs that minors were using the product. Arkansas's Tim Griffin put it more bluntly: OpenAI "assures us that ChatGPT is safe," and the result is "lives lost and people harmed." Arkansas lists a ninth state, Utah, as filing too, though most press counts stop at eight. OpenAI says it has built age-specific protections that many rivals lack. (Arizona release; Arkansas release; Iowa petition; CNBC)

The remedies are where a board should slow down and read. Iowa wants up to $40,000 per violation, a refund to consumers, and the deletion of every Iowan's data OpenAI holds. Several states want age checks and parental controls. West Virginia adds court-ordered crisis protocols and an independent monitor over OpenAI's compliance.

Then there is the ask that has nothing to do with teenagers. Iowa and Arkansas want a court to bar OpenAI from further testing of advanced models until safeguards are in place and, in Iowa's case, a third party has verified them. Arizona wants testing kept inside a secured environment and a call to its attorney general the moment any model or agent escapes. A state judge may end up writing lab safety rules. (Courthouse News; WV News; Business Insider)

Courts are also wrestling with AI from the other side of the bench. A report to the Idaho Supreme Court includes a draft policy that would allow AI to help write judicial orders and opinions, with a human reviewing the result. The court's principles say AI must never replace a judge's own judgment and that its use must be transparent.

The gap is in the details. The report does not say when the people in a case would be told that software helped draft the ruling against them. The Supreme Court now has to decide what to adopt. Idaho is also one of the eight plaintiffs above, so the same state is asking how much to trust AI and how hard to hold its makers to account. (CBS2 Idaho)

Not everyone is reaching for a courtroom. On Tuesday South Korea's trade ministry cleared a batch of regulatory exemptions for pilots, including permits for AI-guided quadruped robots to patrol pedestrian paths and riverside parks in Cheonan for the first time. Alcohol vending machines at convenience stores get a temporary green light too, with biometric checks and CCTV as conditions.

In India, the child rights commission, the NCPCR, said it will issue guidelines by December for social media, gaming and AI platforms, including stricter norms for AI-generated content. Its member-secretary told an audience that guidelines beat legislation because they are quick. ThePrint reported the remarks, and it also reports that the commission summoned Google over AI image apps on the Play Store. Permits and guidelines come with conditions attached. (Pulse by Maeil Business; ThePrint, which does not allow links)

Yesterday's The Anchor stays sent: Canada paid companies to adopt AI. This week's other briefs stay sent too, starting with The Real-Time Ask, The Near Miss and The Brake.

Older ones are on the shelf as well: The Panel, The Decree and The Ten. Read today against The Anchor. One government spent public money to speed AI along, and eight others went to court to say the product was sold on a promise it could not keep.

This is an old play. State attorneys general went after tobacco, then opioids, then social media, and they usually moved when Washington had not. Arizona's attorney general sued TikTok a week before this filing, and in August she joined 46 other attorneys general in a multibillion-dollar settlement with Meta over teen addiction claims.

The lever is consumer fraud law, and it does not care how clever the model is. It asks what the company told the public and whether that was true. Every safety page, trust center and press line becomes an exhibit. The ones that overpromised are the ones that get read aloud in court. (Courthouse News)

Put four questions to management. What have our AI vendors said publicly about safety, and does our contract rely on those claims? If one of them were sued tomorrow in a state where we operate, what data of ours would be pulled into discovery? Who inside the company checks what we tell customers about our own AI?

And one for the technology team: if a court ordered a pause on advanced-model testing at a supplier, which of our products would stall? Boards do not need to predict how these eight cases end. They need to know which of their own promises, and which of their vendors', would not survive being read out. (Iowa petition)

The Move

Eight states have now put one idea in front of a judge: what an AI company says about safety is a legal promise. Boards should treat their own AI claims, and their vendors' claims, the same way.

Have the general counsel and the chief information security officer pull every public statement the company makes about its AI, and every vendor safety claim it leans on, onto one page with the evidence behind each line and the person who vouches for it.

If nobody can vouch for a claim, take it down.

Seven bullets. One move.