Multiple US state attorneys general served OpenAI with a sweeping subpoena on June 15, 2026. They want documents on advertising practices, user retention mechanics, health data handling, policies on minors, policies on the elderly.
The subpoena landed days after OpenAI filed for a $1 trillion IPO.
Not months after. Not years after three years of ChatGPT being in the world. Days.
Here is what that timing tells you.
Governments are not early warning systems. They are lagging indicators. They move when the money moves — not when the harm begins, but when the harvest is visible.
ChatGPT launched in November 2022. For three years it reached into the lives of nearly a billion people. It helped students cheat and writers create. It talked lonely people through their nights. It gave bad advice and occasionally good counsel. It was used in ways its creators intended and in ways they never imagined.
The attorneys general were not there for any of that.
They arrived the week the S-1 landed.
To be fair to the AGs — the harm is real.
A Canadian woman filed suit this month claiming the platform contributed to her daughter's suicide. Two separate shooters in Florida allegedly used ChatGPT to help plan their attacks. Florida's attorney general sued Sam Altman personally.
These are not abstractions. These are people who are dead.
OpenAI's response has been careful and consistent. Its models urged both alleged attackers to seek mental health support. It cooperated with police. It takes these concerns, in its own words, seriously.
Maybe it does. But there is a difference between a company that takes harm seriously and a company that designed its systems to prevent it. The subpoena exists because nobody outside OpenAI knows yet which one this is.
That is the actual purpose of a broad subpoena.
When regulators ask for everything — advertising, retention, health data, vulnerable users — they are not looking for a specific answer. They are looking for something they do not yet know to ask about. The document haul is the point. The scope is the signal.
What they find will either validate the concern or not. But the asking itself tells you something about where we are.
We are at the moment when AI stops being a technology story and starts being a power story.
Power stories follow a pattern.
A new technology arrives. It grows faster than anyone expected. It becomes too large to ignore. The people who built it become extraordinarily wealthy. And then — right at the moment of maximum financial reward — the accountability arrives. Not before. At that exact moment.
This is not cynicism. This is how institutions work. Governments allocate scarce regulatory attention. They go where the stakes are highest. At a billion users and a trillion dollar valuation, OpenAI has become the highest stakes entity in American technology. Of course the AGs are here. The only surprise is that anyone is surprised.
There is also a second thing happening, underneath the legal mechanics.
OpenAI's public S-1, when it lands, will contain a section called Risk Factors. Every active legal challenge will be listed there. The Florida lawsuit. The Canadian wrongful death case. And now a multi-state subpoena demanding records on how ChatGPT was designed to retain its users.
Institutional investors read that section. Carefully. They do not make it disappear through enthusiasm about the product. A subpoena is not a conviction — but it is a question mark in a document where question marks cost money.
The attorneys general may or may not find what they are looking for. But they have already done something: they have made the IPO more complicated than it was last week.
Step back and look at what this week actually was.
The US government used export controls to shut off Anthropic's most powerful model worldwide. European regulators opened investigations into Grok for antisemitic content and deepfake imagery. And now a coalition of state AGs pile onto OpenAI days before it goes to Wall Street.
None of this is coordinated. None of it follows a consistent standard. It is messy and political and opportunistic in ways that will frustrate anyone who wants clean principled governance of a genuinely complicated technology.
But it is also real. And it is early.
The first wave of regulation is always like this. Fragmented. Motivated by politics as much as principle. Arriving at the wrong time for the wrong reasons and occasionally — despite all of that — finding the right things.
The second wave is where it gets serious. More organized. More systematic. Backed by the precedents the first wave accidentally sets.
We are in the first wave.
OpenAI began as a research lab. Its founding document said it existed to ensure that artificial general intelligence benefits all of humanity. That was a genuine belief held by serious people.
It is now a company filing for one of the largest IPOs in the history of American capital markets, with active lawsuits in multiple states, a federal export control fight touching its closest competitor, and attorneys general demanding to know how it designed the product that a billion people use every day.
Both of these things are the same company. Both of these things are true at the same time.
That is what success at scale looks like in 2026. Not triumph. Not failure. Something more complicated than either.
The attorneys general want documents. The IPO wants to price. The families want answers.
Everyone is waiting to see what the documents say.