The Trump administration just dropped a new AI executive order, and it’s exactly as messy as you’d expect from a White House that can’t quite decide how it feels about regulating Technology.
According to reporting from WIRED, the president signed a scaled-back version of an order he had shelved less than two weeks ago after senior aides convinced him they couldn’t keep kicking the can down the road on AI governance. The big change? The original 90-day advance review period that had AI companies collectively freaking out has been trimmed to just 30 days. That’s the compromise that made this thing signable.
Here’s the basic deal: the federal government will get access to the most advanced artificial intelligence models 30 days before they’re released to the public. This gives officials time to poke around for vulnerabilities that could be exploited against critical infrastructure before the models hit the open internet. It’s not formal regulation, it’s a voluntary framework for determining which AI systems are powerful enough to warrant early government review.
The timing matters. We’re talking about increasingly sophisticated systems here, including models like Anthropic’s Claude Mythos and OpenAI’s GPT-5.5. The administration is worried these things could be used to launch cyberattacks against power grids, financial systems, or other sensitive targets. That concern isn’t manufactured. As AI capabilities scale, so does the potential attack surface.
This order is being framed as a victory for White House Chief of Staff Susie Wiles, who worked alongside Treasury Secretary Scott Bessent and National Cyber Director Sean Cairncross to revive the proposal. Notably, this happened despite resistance from Trump’s former AI czar David Sacks, who has been one of the loudest voices skeptical of government intervention in the sector. Sacks was reportedly dialed into the Monday meeting where this all came together, which tells you something about how the internal debates have shifted.
The order also clears the path for Bessent to explore talks with China about creating a similar cross-border framework for advanced AI systems. Those conversations had been on hold while the administration figured out its domestic position. Now that there’s a framework in place, the diplomatic side can move forward.
Some of the big AI players signaled their approval. Anthropic put out a statement calling it “an important step in strengthening America’s leadership in AI” and expressed willingness to collaborate on implementation. That’s the kind of corporate PR speak you’d expect, but it also reflects a genuine reality: companies would rather deal with a predictable 30-day window than live with regulatory uncertainty forever.
What struck me about this story is how clearly it shows the administration wrestling with itself. Trump scrapped an earlier version of this order on May 21 after companies and Sacks warned the 90-day review would be too burdensome for a fast-moving industry. Then executives apparently told the White House that their models were only getting more powerful, meaning kicking the can forever wasn’t really an option. The pressure from the industry shifted from “don’t do this” to “just tell us the rules so we can plan.”
So now we have a framework that gives the government some oversight without strangling innovation, at least according to the White House. The order directs the Pentagon to shore up its classified networks within 30 days and instructs the Justice Department to bring criminal cases against people who use AI models to hack computer systems. That’s the enforcement teeth part of the equation.
It’s a classic Washington outcome: nobody loves it, but everybody can live with it. The question now is whether 30 days is actually enough time to find meaningful vulnerabilities in systems that are becoming exponentially more complex. That seems like a genuine open question, and one we’ll probably get answers to sooner rather than later.
