Everyone in Washington had cleared their calendar for Thursday. The signing ceremony was booked. The photographers were ready. And then Donald Trump walked into the Oval Office and pulled the plug on a widely anticipated executive order that would have required AI models to be evaluated for security risks before public release.

His stated reason: “I didn’t like certain aspects of it.” Also: “We’re leading China, we’re leading everybody, and I don’t want to do anything that’s going to get in the way of that lead.”

That is your AI policy from the White House in 2026. We are winning, regulation might slow us down, therefore no regulation. Simple, elegant, and absolutely maddening if you are one of the cybersecurity researchers who spent the last six months watching AI models get deployed with the ability to find and exploit security vulnerabilities in critical infrastructure.

Because here is the thing the White House is not saying out loud: part of what prompted this order in the first place was the release of Anthropic’s Mythos and OpenAI’s GPT-5.5 Cyber, both of which are capable of rapidly identifying and exploiting software vulnerabilities. These are not hypothetical risks. The models exist. They work. And right now, the only thing standing between a bad actor and a working AI cyberattack is the goodwill of the companies that released the models and the difficulty of gaining access. Those are not the most reassuring guardrails you have ever seen on something this consequential.

An unofficial explanation floating around Washington is even more embarrassing than the official one: not enough tech CEOs could make it to the signing ceremony on short notice, so the event was postponed. This is where American AI governance is in 2026. National security policy on one of the most strategically important technologies in human history, timed around executive scheduling conflicts.

To be fair, the order that was drafted was already watered down considerably from earlier versions. Previous drafts included mandatory model testing before public release. The final version before Trump pulled it was going to task the Office of the National Cyber Director with merely developing a process to evaluate AI models, which is several levels softer than actually requiring that evaluation to happen. Even that was apparently too much oversight for the current moment.

The AI industry breathed a collective sigh of relief, because any regulatory framework creates compliance costs and slows deployment timelines. But it is worth noting the awkward footnote: Microsoft, Google, and xAI had already agreed earlier this month to voluntarily let the U.S. government test their models before launch. The three biggest players in American AI had already consented to exactly the kind of oversight that Trump just decided not to require. Make of that what you will.

The order will presumably be reworked, reworded, and signed at some future date when the language is more to everyone’s liking and the right tech CEOs can get a flight to D.C. Until then, America’s official posture on AI security is: we are thinking about it, we just do not want to think too hard right now, because we are very busy winning.

Keywords: Trump AI executive order, AI safety regulation, AI cybersecurity policy, White House AI policy, OpenAI GPT-5.5 Cyber