Trump administration weighs AI model reviews as tech giants race to ship faster

The Trump administration is looking at a new order that would let US security agencies check powerful AI models before companies put them out for the public.

The plan came up in a White House briefing led by the Office of the National Cyber Director. The meeting included OpenAI, Anthropic, and Reflection AI, all private companies, so there are no stock tickers for them.

The order could be signed by Donald Trump as soon as Thursday. After the first mention, Trump is the name used here. The plan would set up a “voluntary framework” for companies building frontier AI systems.

Under that setup, AI firms would tell the US government before major launches. They could also give agencies access to advanced models up to 90 days before those models reach users.

Trump lets agencies check frontier AI systems before public launches

The AI section of the order will emphasize “covered frontier models,” which means that the government will first determine which AI technologies are sufficiently significant to receive additional review. This won’t involve an examination in one particular office but rather several agencies who will evaluate models before their release.

The expected executive order will have two key provisions, one related to cybersecurity and another one related to advanced AI models.

While the cybersecurity provision will target the Pentagon, national security agencies, hospitals, financial institutions, and other critical infrastructure throughout the country, the second provision is related to expanding the pool of cyber experts hired.

This includes increasing the number of employees at the US Tech Force – a program launched by OPM director Scott Kupor late last year. According to Scott Kupor, the purpose of the US Tech Force program was to recruit top-tier AI specialists in federal agencies.

The order would also push AI companies and the government to share more details about security breaches. That part is about speed. If a company finds a weak spot or gets hit, federal teams want that information faster, not three meetings and a dead inbox later.

Treasury builds a clearinghouse while NSA gets final AI review power

The Treasury Department would lead a voluntary project with AI companies and owners of critical infrastructure. That project would create a clearinghouse within 30 days. The job of the clearinghouse would be simple: find security holes and help fix them.

The Office of the National Cyber Director, the National Security Agency, and the Cybersecurity and Infrastructure Security Agency would support Treasury’s work. CISA and the National Institute of Standards and Technology would also help build the model review process.

The second section would give Treasury, CISA, and NIST 60 days to create a classified test process for deciding what counts as a covered frontier model.

White House chief of staff Susie Wiles, National Cyber Director Sean Cairncross, and Michael Kratsios, who leads the White House Office of Science and Technology Policy, would also take part. After that, Susie, Sean, and Michael would stay tied to the process through their offices.

The NSA would have the final say after speaking with the other agencies. The White House started meeting with tech and cyber groups after Anthropic showed Mythos last month to a small group of tech companies and security researchers.

A White House official called the reports “speculation” and said any real announcement would come from Trump. The drafting has also shown disagreements inside the Trump administration over how much review frontier AI models should face before launch.