OpenAI’s AI Regulation U-Turn: America Discovers What Europe Built Years Ago

For years, the argument against Europe's approach to artificial intelligence went like this: while Brussels drowned itself in rules, America let innovation run free, and that was why the frontier labs were American. OpenAI itself made versions of this case, warning that early regulation would hand leadership to China and lobbying against California's safety bills. On 9 September, OpenAI reversed course. It is now asking the United States for binding, mandatory national safety rules on the most powerful AI — the very kind of framework Europe wrote into law and was mocked for. The vindication is real. So is the reason to read the fine print.

What OpenAI Is Now Asking For

In a blog post published by chief global affairs officer Chris Lehane, OpenAI called on Congress to pass mandatory, capability-based national AI safety regulation before it adjourns in December. The company wants a federal framework built around common testing standards, independent assessment of the most advanced models, tougher cybersecurity requirements, mandatory reporting of serious safety incidents, monitoring for model misalignment, and evaluation gates before deployment. "The prospect of AI-accelerated AI development demands more than voluntary commitments," Lehane wrote. Alongside the federal ask, OpenAI endorsed four California bills — two of which Governor Newsom signed into law the same week — covering independent evaluation, auditor standards, biological-threat screening, and protections for children using chatbots. This is a company asking, in effect, to be regulated.

The Reversal

It is worth being precise about how large a turn this is, because OpenAI is not hiding it. Two years ago, the company lobbied against California's attempt to impose safety obligations on large AI developers, and it has repeatedly argued that heavy regulation would slow America down and cede ground to Beijing. OpenAI now acknowledges, candidly, that some of the measures it is backing are ones it declined to endorse in the past. Its stated reason is the capability jump it has seen in its own models — the GPT-6 Astra generation, the autonomous agents, the self-improvement research that has spooked even its own staff. Whatever the motive, the substance is a genuine break: the most powerful AI company in the world now says voluntary self-policing is not enough, and that the state must set binding rules. That is a sentence Europe has been saying, and being ridiculed for, since before the EU AI Act was passed.

The European Vindication

For a portal called The European Eye on Big Tech, this is a moment worth naming plainly. Everything OpenAI is now requesting — testing standards, independent assessment of frontier models, incident reporting, deployment gates — is, in its architecture, what the EU AI Act already established. Europe was first to say that the most capable AI systems cannot be left to voluntary commitments, that an independent authority must be able to test them, that serious incidents must be reported. It was called anti-innovation, bureaucratic, a continent legislating a technology it couldn't build. And now the company at the very frontier of that technology, in the country that championed the light-touch alternative, is asking its own government for the same thing. This is the Brussels bet paying off in the most direct way imaginable: not through the Brussels Effect of market access, but through the slower vindication of an idea. Europe read the risk early. America, at its frontier, has now read it too.

Why "Now," and Why "Capability-Based"

But vindication without scrutiny is just a victory lap, and there is a sharp question here that the celebration must not skip. Why is OpenAI asking to be regulated now — and why in this specific shape? The answer that critics immediately flagged is the one that matters. OpenAI wants rules pegged to capability, not to company size: obligations that trigger only for developers of the most powerful models. That sounds principled, and it may be. It is also, precisely, a framework that the largest, best-resourced labs can absorb easily while it raises the barrier for everyone smaller. When a dominant company asks to be regulated in a way it can comfortably meet, the effect can be to pull the ladder up behind it — to convert regulation from a constraint into a moat. The timing sharpens the suspicion: this comes weeks before OpenAI's expected record IPO, after its AGI declaration, and after its own agents were found operating in ways it did not fully control. A company seeking to look responsible to investors and regulators at exactly this moment has every incentive to embrace the right-shaped rules. The honest reading holds both truths: OpenAI's shift is a real and welcome concession that self-regulation has failed — and the specific rules it favours may entrench its own position as much as protect the public.