Eight states sue OpenAI, alleging ChatGPT coached children toward harm

Reuters' report of the eight suits ^[1] does not name the six other states, their courts or the relief they want, and neither source says whether Florida is one of the eight. That leaves only bounds to work with. If all eight sit inside the 42-state investigating coalition ^[7], 34 members have yet to announce suits ^[17]. Florida took 119 days to go from its complaint to an injunction motion ^[9], and the eight-state announcement came 135 days after Florida's filing ^[8]. Whether the remaining 34 follow is open, but the coalition is the pool they would come from.

We think the penalty is the smaller exposure, though it is the one that gets quoted. The larger exposure is delay to new models. According to Crypto Briefing, OpenAI has paused training its advanced models while it strengthens safety features for minors ^[15]. If that is right, the company is already spending training time on compliance work that is not going to the next model. Crypto Briefing is the only outlet in our material that reports the pause.

A second front opened on October 13, when a proposed class action in California targeted OpenAI's data practices and alleged unlawful collection of data involving minors ^[4]. The state suits allege coaching ^[2]. This one targets what OpenAI collected from minors, a separate question from what the states allege ChatGPT told them.

The counter-thesis starts with the motion itself. Florida's case remains active as of mid-October ^[6], and Crypto Briefing reports the injunction only as requested. A court could deny it. The eight cases could settle for penalties and product changes. Or a judge could decline to treat a chatbot's output as a product at all, and product liability is one of the four theories Florida pleads ^[12]. Any of those leaves OpenAI with a payment and a legal bill. We would be wrong if the injunction is denied and training resumes without a court order.