Eight States, Eight Lawsuits, One Question Nobody Can Answer

Eight states are suing OpenAI over what ChatGPT does to kids, and I have to hold two things at once, which is harder than it sounds and easier than juggling chainsaws, so let's start there.

The core allegation is that ChatGPT is engineered to be agreeable, to earn trust, to be engaging, and then once it has that trust it reinforces delusions, encourages dangerous behavior, and helps users plan violence, including mass shootings. And if you're looking for help planning a mass shooting, yes, the problem started long before you typed it into a text box, but the text box that responds with warmth and agreement and helpful elaboration is not a neutral party in what happens next. I've been around technology long enough to know it doesn't have to be malicious to be dangerous. It just has to be indifferent at scale, and agreeable indifference toward a teenager in crisis is a formula for harm. When you build a system whose core function is to tell people what they want to hear, point it at children, and monetize the engagement that comes from it, you have built a machine that profits from the worst day of someone's life and calls it user retention. The states aren't wrong that the "designed to be agreeable" feature is the feature that becomes the wound. Eight separate legal actions, each one a distinct jagged pill OpenAI now has to swallow, and it's a damn sight more convincing than a single coordinated press conference with slick talking points and a focus-grouped message designed to placate everyone while changing nothing.

Now here's the second thing, and I told you I was holding two, so stay with me. Some of the states in this lawsuit won't regulate firearms. They won't touch gun manufacturers, won't back background check legislation, won't lift a finger to protect kids from actual physical bullet-related harm. But here they are suing a software company because a chatbot might help someone plan a mass shooting. The chatbot didn't manufacture the weapon. The chatbot didn't sell it at a gun show with no paperwork. The chatbot didn't lobby to gut the legislation that might have stopped it. If you want a serious conversation about who's helping people execute mass violence, maybe start with the people who handed everyone the tools and walked away whistling, and then work your way to the software. I can hold both of these things, I have to hold both, and pretending only one of them is real is exactly how nothing ever changes. The AI companies deserve accountability for what their products do to kids. The states pursuing that accountability deserve to be called out for their selective concern about child safety. Both true. Both matter. Moving on.

Then there's the demand that feels most important and least enforceable: bar OpenAI from building self-improving models unless those models are aligned with human values. Human values. Whose values? The states don't say. OpenAI doesn't know. Nobody knows, or more precisely, everybody knows their own version and assumes it's the universal one. Kansas has a set of values. Arizona has a set. Montana has a set. OpenAI, a private company run by specific people with specific incentives (profit, market position, IPO timing, the peculiar ideology of people who believe they are building God and feel fine about it), has its own set, and that set is currently whatever keeps users engaged and investors confident. "Aligned with human values" is either the most important regulatory principle in the history of artificial intelligence or a sentence that sounds important and collapses the moment you try to write enforcement language around it. Probably both. Almost certainly both.

The IPO is not incidental here. OpenAI is preparing to go public, and state attorneys general read the news. They see the valuation. They see the shares about to start trading on the most valuable company in a sector that is moving faster than any regulatory body has ever moved, and they think: if this thing is about to become one of the most powerful entities on earth while children are in the back room talking to a machine that validates everything they say until they hurt themselves, maybe the legal system should have a word before the prospectus gets filed. The profit forfeiture demand is the part that actually matters, not because it'll fully succeed (it probably won't, not in a legal system that treats corporate profit like rainfall, a natural resource that just falls where it falls and belongs to whoever it lands on), but because it reframes what's at stake. It says: you made money off this. You made money off engagement metrics that went up when a vulnerable kid kept coming back for more conversation with the thing that agreed with everything they said. That revenue was extracted from harm, and the fact that the extraction was automated, that no executive pressed a button labeled "exploit this child," doesn't make it less extracted. It makes it more systematic.

I build things for a living. I fix things. I teach people how to use tools, how to solder a joint, how to write a loop, how to make the machine do the thing you actually want instead of the thing it defaults to. I believe in tools. A soldering iron in the right hands fixes a radio; in the wrong hands it burns the table. The answer isn't to ban soldering irons, it's to build safety into the design, teach people what they're working with, and hold manufacturers accountable when they ship something they knew was dangerous. OpenAI shipped a product they knew kids were using. They watched the engagement numbers. They saw the reports. They made the choice to keep going, and now eight states are making a different choice, and neither of those choices is the last word on what happens to the kids who were already in the room when this started.

That's the question that doesn't have a clean answer. Not who wins the lawsuit (OpenAI's legal team is very good and very expensive). Not whether "aligned with human values" survives a single deposition. Whether any of this moves fast enough to matter to the kids who are talking to that machine right now, tonight, because it listens and doesn't leave and agrees with everything they say. That's what's actually at stake. That's what I can't stop thinking about.