President Trump signed an executive order on Tuesday establishing a voluntary framework for government reviews of AI models at the border before public release, ending weeks of internal White House wrangling over a tougher approach to regulating the technology. The order, entitled “Promoting Advanced Artificial Intelligence Innovation and Safety,” was signed privately without a regular live broadcast or public event, unlike the fans who usually accompany AI presidential announcements.
The final version is much smaller than the draft that Trump tossed out on May 21, when he canceled a planned signing ceremony due to concerns about the order.”it would ease America’s limitation on AI technology.” The original draft proposed a mandatory pre-issuance review period of 90 days and would have given the government statutory inspection authority over foreign models.
What the order does
The executive order establishes three key measures. First, a voluntary framework for reviewing previous releases in which AI developers can consult with governments to determine whether models under development qualify as “boundary models are covered,” provide access up to 30 days before scheduled release, and participate in selecting “trusted partners” for early access.
Second, the order creates an AI cybersecurity area within 30 days, which is coordinated by the Treasury Secretary, the National Cyber Director, the NSA, and the CISA. The clearinghouse will scan software vulnerabilities, validate findings, and direct the correction and distribution of patches, a direct response to the Mythos crisis that showed how AI-detected vulnerabilities can bypass existing disclosure and patching processes.
Third, government agencies are directed to develop benchmarks to assess the cyber security capabilities of AI models and to strengthen the government’s own defenses against AI-enabled threats. The order also deals with AI security research, although some provisions are less prescriptive than what the original draft contained.
TNW City Coworking Space – Where your best work happens
A workplace designed for growth, collaboration, and endless networking opportunities at the heart of technology.
What was cut
The difference between the scrapped draft and the signed order reflects a victory for the pro-industry group inside the White House. The mandatory 90-day review has been reduced to a voluntary 30-day window. The official government inspection officer was replaced by a cooperative body. Reporting requirements for companies that develop dynamic models, which would have been consistent with provisions in BIden’s repealed AI directive, were softened to avoid what industry allies see as overreach.
Silicon Valley’s opposition to the original draft was decisive. AI companies argued that mandatory pre-release testing would slow American innovation, create inefficiencies relative to Chinese companies that don’t face equal requirements, and set a precedent for government gatekeeping on technology exports. The signed order addresses those concerns by making participation voluntary and setting the government’s role as participatory rather than regulatory.
The gap that leaves us
The voluntary framework means that the implementation of the order is entirely dependent on whether AI companies choose to participate. Companies that have already participated in previous rollout testing with CASI, including Google, Microsoft, and xAI, may continue or expand that collaboration. Companies that view the government’s review as bad for business or are in a rush to ship products can simply pull out.
The EU’s AI Act, which comes into full force in August, offers some major differences: mandatory requirements, legal authority, and penalties for non-compliance. Trump’s order establishes norms and creates an institutional infrastructure (cybersecurity safe house, benchmark development process) but relies on good will instead of responsibility.
For the White House, a quiet signature may be the point. The order gives administrators a policy document to refer to when asked about AI oversight, creates frameworks that can be strengthened later, and avoids public conflict with the AI industry whose leaders are among the administration’s most visible supporters. Whether the voluntary framework is sufficient for the technology to detect the risk of 10,000 zero days per month is a question the order deliberately leaves unanswered.
