US presses Meta to agree with AI reviews

The Trump administration is urging Meta to submit its AI models to voluntary review, which would allow the government to assess AI capabilities and vulnerabilities, four people familiar with the confidential request said.

The request, which was sent in emails to Meta, is the latest example of the administration’s push to strengthen oversight of the artificial intelligence industry after pushing for unrestricted access to the powerful technology. Less than two weeks ago, the government ordered Anthropic to revoke access to its latest model, citing national security concerns.

Meta is the only major U.S. AI developer that has not reached an agreement to voluntarily share its models with the federal government for review, said people familiar with the request, who spoke on condition of anonymity because they were not authorized to discuss the matter publicly. OpenAI, Anthropic, Google, xAI and Microsoft have all agreed to submit their models to the government’s AI safety group, known as the Center for AI Standards and Innovation.

“We share the administration’s goal of advancing U.S. leadership in robust and secure border AI,” Meta spokesman Francis Brennan said in a statement Tuesday. “While we are working on the details, we hope to sign the contract soon.”

Ben Kass, a spokesman for the Commerce Department, said the department’s Center for AI Standards and Innovation, which is based in the department, regularly talks with companies about voluntary agreements.

“This story is not unusual,” he said. “This is exactly the job CAISI is supposed to be doing.

The latest batch of artificial intelligence models has raised concerns about cybersecurity, though some industry insiders say the concerns are overblown because the new technology can be used to defend computer networks just as easily as it can be used to attack them.

The companies have been submitting their AI models for review for several months as a goodwill gesture so intelligence and defense officials could ensure that the latest AI products do not pose a national security risk, two government officials said.

On June 2, President Trump signed an executive order giving the government responsibility for AI controls. The order required tech companies to give the U.S. government up to 30 days to evaluate AI models before making them public. It gave the government until the end of July to develop a review process.

But it’s unclear who will lead the effort and what type of standards the models should be held to, said one person involved in the process.

The Center for AI Standards and Innovation has stepped up its efforts to play a role in the model review process. The agency, overseen by Commerce Secretary Howard Lutnick, was created by the Biden administration to vet AI models and has a technical staff to conduct those evaluations.

Meta released its latest AI model, Muse Spark, in April, and the performance nearly matched models from competitors such as Google, OpenAI, and Anthropic.

Meta’s core team has been negotiating with the sales department about how to proceed, people familiar with the confidential request said. It is not clear whether they will be able to come to an agreement.

Even companies that have given the administration previews of their AI models have run into trouble. Anthropic, which provided its latest and most powerful model, the Fable 5, to the administration for review last week when the White House gave the company less than 90 minutes to shut down access to its new AI over national security concerns.

The problem, said two people with knowledge of the discussions, stems from a paper written by Amazon researchers that showed a vulnerability in the model that could be exploited for cyberattacks.

In the days that followed, talks between Anthropic and the administration were productive, said two people with knowledge of the talks. Mr Trump made the comments over the weekend to Axios that he no longer saw Anthropic as a security concern.

It was unclear who would sign off on renewing Anthropic’s access to its model and whether other companies would be held to the same standard, the two people said.

Ana Swanson contributed reporting from Washington.