Trump Eyes Federal Reviews for Advanced AI Models Before Launch

Washington may be getting nervous about what the next frontier AI model can do before the public ever sees it.

President Donald Trump’s administration is considering federal reviews for advanced AI systems before they launch, according to The New York Times. The possible process would focus on models whose capabilities could affect cybersecurity, national security, or military use, though it may not give the government authority to stop a launch.

For an administration that has favored fewer restraints on Silicon Valley, the idea creates a new test: how to keep the AI race moving without letting the most capable systems become a security risk.

Why one unreleased AI model raised alarms

The concern escalated after Anthropic developed Mythos, a model it described as powerful at identifying software security vulnerabilities. The AI company said the model could lead to a cybersecurity “reckoning” and declined to release it publicly.

That decision turned a policy debate into a launch problem. A model capable of surfacing software flaws at that level could become a powerful tool in the wrong hands if released without guardrails or government visibility.

Federal officials are also looking at the upside. AI models like Mythos may have value for the Pentagon and US intelligence agencies, particularly if they can help identify weaknesses in government systems before adversaries exploit them.

What a pre-launch AI review could include 

One version under discussion would give federal officials access to new AI systems before they are publicly available. People briefed on the talks said the review process would not necessarily give the government the power to block a launch, limiting its role to early review rather than final approval. 

The agency lineup is still unsettled. Officials have discussed roles for the National Security Agency, the White House Office of the National Cyber Director, and the Director of National Intelligence. The Center for AI Standards and Innovation, a Biden-era office created to assess voluntarily shared AI models, could also be brought back into the process after being sidelined under the Trump administration.

The security test for Trump’s AI policy

The discussions would test an AI agenda built around speed, competition, and fewer federal restraints. Since returning to office, Trump has treated AI development as central to the race with China and moved quickly to unwind Biden-era requirements that asked developers to run safety evaluations and report on models with potential military uses.

The possible review plan runs counter to that record. After warning against rules that could slow AI development, the administration is now weighing a federal checkpoint for its most powerful systems.

Tech companies split over AI oversight and China

Silicon Valley has its own warning for Washington. Move too far, and the US could slow itself down against China.

Executives are also split over what federal rules should look like for a review process that could give the government an early look without giving it final authority over a launch.

“The technology is moving extremely fast, and there are few formal procedures, but they also don’t want to overregulate,” Dean Ball, a former Trump administration AI adviser, told the Times.

The unresolved question is whether Washington can build a review process that catches dangerous capabilities early without turning every major AI launch into a federal approval fight. For now, the talks suggest that even an AI policy built around speed may need to accommodate security checks once the models become powerful enough.

A Chinese court has made clear that replacing workers with AI may not be as legally simple as companies hope.