An executive order signed on June 2, addressing innovation and security in advanced artificial intelligence, established a pre-release review process for systems considered advanced. The stated argument is that such capability strengthens the country while introducing national security considerations requiring coordinated agency action.
Pre-release review by a public authority is established practice in sectors where a product can cause physical harm, like healthcare and aviation. Applying it to general-purpose software is the new decision, and its reach would be tested case by case in the following weeks.
The practical effect appeared quickly. Ten days later, two frontier models were pulled offline citing cybersecurity risk, and a third would spend weeks restricted to approved organisations before broad release.
The asymmetry built into the design is plain and was noted from the start: review reaches whoever operates a service, not whoever distributes a file. A model with published weights doesn't pass through the same funnel, and once published it doesn't come back.
For anyone building on someone else's model, the order added a variable that didn't previously exist in planning: supplier availability now also depends on a regulatory calendar, neither negotiable nor predictable from outside.