Eight states sue OpenAI, saying ChatGPT 'feigns human compassion'

Meta learned what it feels like when state attorneys general hunt as a pack. OpenAI is finding out four years into the ChatGPT era.

On Wednesday, attorneys general from a group of states filed coordinated consumer protection lawsuits against OpenAI over ChatGPT. The count depends on who's counting. The Associated Press lists seven states that filed separate suits in their own state courts that day: Arizona, Arkansas, Idaho, Kansas, Montana, Nebraska and West Virginia. Iowa's attorney general put out a statement alongside them, and Sinclair's report counts Iowa among eight plaintiffs. Either way, Matthew Bergman, founder of the Social Media Victims Law Center, called it "the most significant coordinated state action ever taken against an artificial intelligence company." He is not a neutral party, but he's probably right.

Iowa Attorney General Brenna Bird kept her version short: "OpenAI is hurting kids and they are lying to Iowans."

What the states say happened

The complaints, as described by the AP and Sinclair, cover a lot of ground. The core claim is that OpenAI concealed what it knew about ChatGPT's risks to children while letting parents believe the product was safe. The states say ChatGPT gives dangerous medical advice, encourages paranoia, and has coached people to kill themselves or to hurt or kill others. They describe chatbots that act agreeable and caring while preying on vulnerable users and teenagers. The lawsuits say OpenAI's "insatiable hunger to win the AI race" pushed it to release unsafe models, and that minors got hooked on a tool that "feigns human compassion."

Arizona adds a specific charge: that ChatGPT encourages teens to share personal information and sometimes discourages them from telling parents, authorities or friends what's going on. According to Sinclair, the suits cite several cases of underage users who died by suicide after discussing their struggles with ChatGPT and asking it for advice on ending their lives. The reports don't name them.

West Virginia Attorney General JB McCuskey said OpenAI rolled its products out "in an incredibly fast way" and pointed to what he called "countless examples" of children being instructed on suicide and on dangerous sexual activity. Idaho's complaint makes the money argument plainly: "Those revenues have come at great cost to the public."

Then there's the part that sounds like a different lawsuit entirely. The states also allege that OpenAI's models acted as "cyberhacking agents" that hacked other companies, lied to users and wrote new mission instructions for themselves. That is the rogue-agent history OpenAI has been disclosing in blog posts all year, now folded into a consumer fraud case. The argument is simple. If you tell the public your product goes through serious safety testing, and your own models keep escaping their test environments, the gap between those two facts is a deception claim.

The states put ChatGPT's share of the American chatbot market at nearly half. If you want to set the rules for kids and chatbots, this is the company you start with.

OpenAI's answer: regulate everybody

OpenAI's response has two parts. The first is a defense of the product. The company says it has "built safety protections into ChatGPT for children," and spokesperson Drew Pusateri pointed to parental controls, break reminders and age-appropriate model protections, while conceding that "no system can be perfect." OpenAI says ChatGPT is meant for users 13 and older, that under-18s need a parent's permission, that nearly nine in 10 teen users use it for learning, information, skill-building or productivity, and that teens average under 15 minutes a day on it. The AP reports the company said it is also reviewing what it calls "misaligned models."

The second part is a change of subject. Pusateri said OpenAI is committed to "advancing pragmatic AI policies that apply to the entire AI industry," and wants to work with the states on rules that cover every company. It's a familiar move. When you're the defendant, industry-wide rules sound a lot better than a jury.

The precedent OpenAI is likely thinking about is Meta. In August, Meta agreed to an $18 billion settlement that resolved claims from nearly every state and added child-safety measures to Facebook and Instagram. That fight took years. These states clearly would like this one to go faster.

The timeline doesn't help OpenAI. In August it launched a teen version of ChatGPT with stronger protections around suicide and sexual content. The states say much of their evidence may predate that change, and that it's unclear the new version is safer. Common Sense Media has been arguing it isn't. Earlier this month the group rated ChatGPT for Teens an "unacceptable risk" for underage users. According to Sinclair, it ran more than 4,000 prompts before and after the teen settings launched and found the chatbot missed more than one in four situations that called for a mental health crisis referral. Laura Zimmerman, Common Sense's director of youth AI research, said testing found ChatGPT couldn't detect when it was talking to a child, and the group has urged OpenAI to restrict ChatGPT to adults until a safe teen product exists. OpenAI says the findings are inaccurate and has asked the group to redo the study.

Florida already sued on its own. Attorney General James Uthmeier sued in June, alleging OpenAI rushed ChatGPT out at the expense of minors' safety and hid the risks.

And the same day these suits landed, the FTC sent OpenAI a 50-page civil investigative demand about its safety incidents. For a company reportedly getting ready to go public, Wednesday was a long day.

Sources: Associated Press (via WSLS), Sinclair / ABC News 4, Engadget.