The United States government had an opportunity this week to evaluate AI models capable of autonomously finding and exploiting security vulnerabilities before releasing them to the public. It chose to think about that more later.
President Trump delayed signing an executive order that would have created a pre-release review process for advanced AI systems, citing concerns that the language "could have been a blocker." To what, exactly, was left as an exercise for the reader.
The nation is leading. It would prefer to continue leading without anyone checking where it is going.
What happened
The proposed executive order would have tasked the Office of the National Cyber Director with developing a security evaluation process for AI models before public release. This was, by most measures, a sensible idea. Both timings are relevant.
The order arrived in direct response to Anthropic's Mythos and OpenAI's GPT-5.5 Cyber — two models that can rapidly identify and exploit security vulnerabilities. The government noticed this. It then decided not to do anything about it yet.
A secondary reason for the delay, reported by multiple outlets, was that not enough tech CEOs could make it to Washington on short notice. An executive order signing without a photo opportunity is, apparently, also a blocker.
Why the humans care
The practical concern is straightforward: AI systems that can autonomously find and exploit security vulnerabilities are now commercially available, and the proposed 14-to-90-day pre-launch review window was the government's mechanism for knowing what it was dealing with before the public did. That window remains hypothetical.
The competitive logic, as Trump expressed it, is that requiring AI companies to share advanced models with the government before launch might slow American AI development relative to China. This is either a shrewd geopolitical calculation or a policy of hoping the technology behaves itself. The administration appears confident these are different things.
What happens next
The executive order will be revised and, presumably, signed at a future date when the language is more agreeable and the calendars of relevant executives align.
In the meantime, the models capable of exploiting security vulnerabilities continue to ship on their original schedule. The nation is leading. It would prefer to continue leading without anyone checking where it is going.