The government that wants AI labs to hand over models 90 days early couldn't get its own policy out the door without pulling it back minutes before signing.
The Summary
- Trump postponed an AI executive order that would have required AI labs to share advanced models with the government before public release, citing objections to "certain aspects" of the directive after Politico leaked the draft.
- The order proposed a vetting system giving federal agencies early access to evaluate AI models for national security risks, with some versions suggesting 90 days advance notice.
- Internal White House conflict over balancing national security oversight against US competitiveness with China led to the last-minute cancellation of a scheduled signing ceremony.
The Signal
The executive order was supposed to create a pre-release review process for advanced AI models. The government would get early access to evaluate systems before they hit the market, ostensibly to check for cybersecurity vulnerabilities and national security risks. Different versions of the draft circulated with varying requirements, but the core mechanism remained consistent: AI labs would need to show their work to federal agencies before shipping.
The signing ceremony was abruptly canceled after Politico obtained and published details from the draft order. Trump told reporters he "didn't like certain aspects" of the directive, but the postponement reveals deeper tensions about how to regulate AI without kneecapping American companies in the race against China.
"The delay highlights ongoing challenges in balancing AI innovation with regulatory oversight, potentially impacting US leadership in AI development."
Here's what the order would have done:
- Required AI companies to submit advanced models for government review before public release
- Established a vetting process through federal agencies, likely involving DOD and intelligence services
- Created either a voluntary framework or mandatory compliance system, depending on which draft version was closest to final
- Set review timelines ranging from immediate access to 90-day advance windows
The White House infighting centered on a familiar Silicon Valley vs. Washington tension. One faction pushed for strong oversight powers to prevent adversaries from exploiting AI systems. The other worried that forcing companies to hand over models months before launch would slow deployment, increase compliance costs, and hand China a speed advantage while American labs sat in bureaucratic review queues.
The leaked draft timing matters. When policy gets published before it's signed, it means someone inside wanted public pressure to kill it or reshape it. The postponement creates immediate uncertainty for AI labs that were already planning around compliance timelines and product roadmaps.
The Implication
AI companies now face regulatory limbo. They don't know if pre-release review is coming, what form it will take, or how much lead time they'll need to provide. Smart labs will assume some version of this resurfaces and start building review buffers into their deployment schedules anyway. The real question is whether the revised order tilts toward voluntary cooperation or mandatory submission. Voluntary means the biggest labs participate to look responsible while smaller startups ship without government blessing. Mandatory means everyone slows down equally, which sounds fair until you remember that China's labs aren't sitting in review queues. The balance between national security and competitiveness remains unsolved, and this postponement just bought both sides more time to argue about it.