As President Donald Trump prepares to travel to a summit in China, his administration is sharply split over a plan to give U.S. intelligence agencies a bigger role in evaluating AI models, according to two people familiar with the matter, who spoke on the condition of anonymity to discuss a proposal that is not yet public.
Trump could sign an executive order addressing AI security as soon as Monday. But the administration’s response to new advanced AI models remains in flux and a topic of extensive debate, according to multiple people with knowledge of the conversations.
The debate within the administration pits Commerce Department officials against national security aides in a battle, which one person described as a “knife fight,” to determine which part of the government will have sway over technology that Silicon Valley leaders say can transform the economy.
The debate has taken on greater urgency as Trump prepares for his meeting with Chinese President Xi Jinping and as U.S. companies brace for Chinese competitors to release models with capabilities similar to Anthropic’s Mythos and other advanced artificial intelligence models.
The White House in recent weeks has grappled with new cybersecurity threats posed by Mythos and similarly advanced models. Anthropic has said that Mythos is more effective than humans at identifying vulnerabilities in software, creating widespread concerns that it could be used by even inexperienced hackers to launch complex attacks on the energy grid, banks and government agencies.
In response, the White House’s Office of the National Cyber Director has proposed developing a large center within the Office of the Director of National Intelligence that would evaluate new AI models, giving intelligence agencies a significant new role in AI policy.
That proposal has faced opposition from officials at the Commerce Department, which houses an AI center that evaluates new models through voluntary agreements with companies, according to some of the people who spoke to The Washington Post.
“There is no daylight between the Department of Commerce and this administration on AI policy,” the department said in a statement to The Post. “From export controls to the AI exports program, we are laser focused on ensuring American leadership.”
The ODNI referred inquiries to the White House.
The Commerce Department’s AI center has built an evaluation system and hired AI experts with advanced degrees, said one person, and some in the agency have argued that infrastructure makes the agency better positioned to lead on testing. But the intelligence community has a deep bench of technical personnel with cybersecurity and AI expertise.
The Commerce Department, especially under Trump, has promoted a pro-industry strategy as it tries to ensure that American companies can remain competitive with China. Shifting testing to the intelligence community could lead to greater scrutiny of the cybersecurity and national security risks that new AI models present.
Anthropic’s announcement of Mythos has forced some in the administration to reevaluate the laissez-faire approach they have taken toward AI regulation, which included rolling back many of the measures President Joe Biden’s administration put in place to oversee AI development.
“They’re relitigating everything on AI policy right now,” said Chris McGuire, senior fellow at the Council on Foreign Relations and a senior technology policy aide at the National Security Council during the Biden administration. “Is it voluntary testing? Mandatory testing? Voluntary limits on what’s released? Mandatory limits?”
The White House continues to engage across the government and industry, and top administration officials are discussing approaches with frontier labs, said a White House official, who spoke on the condition of anonymity to describe the private deliberations.
“Any policy announcement will come directly from the president,” the official said. “Discussion about potential executive orders is speculation.”
In a sign of the tension, the Commerce Department’s Center for AI Standards and Innovation on Friday took down a website that announced the agency’s new partnerships to test models from Microsoft, Google and xAI. The website was removed because of sensitivity within the White House’s Office of the National Cyber Director, according to two people familiar with the decision.
As of Monday morning, the website was still down. “Oops, that’s not standard?!” said the website error page. “The page you requested cannot be found at this time. It may be temporarily unavailable or it may have been removed or relocated.”
The announcement previously said the center would “conduct pre-deployment evaluations and targeted research to better assess frontier AI capabilities and advance the state of AI security.”
Under Biden’s AI executive order, AI developers were required to notify the government when they were building advanced models and share the results of testing with the Commerce Department. The Biden administration also established the U.S. AI Safety Institute under Commerce’s National Institute of Standards and Technology, which was part of a global network of centers that evaluated AI models. In a sign of the new administration’s shift, Trump removed “Safety” from the agency’s name and rebranded it as the Center for AI Standards and Innovation.
Administration officials have been divided over whether evaluations of AI models should be mandatory or whether companies could participate on a voluntary basis, as they do now with the Center for AI Standards and Innovation.
“That’s very much a topic of active debate,” said a person familiar with the matter. “There are those in the intelligence community and elsewhere who think it needs to be mandatory.”
The form that evaluations would take if mandatory is unclear, the person said. Among the open questions is whether the administration should pursue a legal mechanism or “just strong-arm companies the same way they do on other things.”
Commerce’s CAISI has a documented record of vetting large language models, the technology powering ChatGPT and Mythos. Earlier this month, the center published its evaluation of the latest version of DeepSeek, an AI model developed by a Chinese start-up that sent U.S. tech stocks tumbling when it debuted in January 2025. In December, CAISI published an evaluation of another AI model from the Chinese start-up Moonshot AI. During the Biden administration, the center also published evaluations of cutting-edge AI models from Anthropic and OpenAI before they were deployed.
In recent weeks, White House officials have put out mixed messages as they consider a reseton their AI policies.
National Economic Council Director Kevin Hassett last week suggested that the administration would evaluate new AI models the way the Food and Drug Administration tests new drugs “so that they’re released to the wild after they’ve been proven safe.”
Hassett’s comments sparked widespread consternation in the industry, as they appeared to be a complete reversal of the Trump administration’s approach to AI regulation.
Within hours, White House Chief of Staff Susie Wiles, who seldom posts on social media, said on X that the president was not “in the business of picking winners and losers.”
Trump’s Silicon Valley allies sought to cast Wiles’s tweet as a rejection of the FDA model. David Sacks, who previously served as the White House’s AI czar and remains active in discussions about how the administration should respond to AI advances, said on the “All-In” podcast, which is widely followed in Silicon Valley, that no senior administration official backs having an FDA for AI models.
Sacks called it imperative to get advanced AI models into the hands of cybersecurity companies to help businesses and the government prepare for the public release of Mythos and other advanced models later this year.
He rejected calls for the government to approve the release of AI models, saying that companies are already acting sensibly.
“Both Anthropic and OpenAI acted responsibly here. No one was trying to release these super powerful models,” he said on the podcast, referring to the ChatGPT maker. “In a way, all these people who are saying we need prerelease approvals for models, they’re trying to solve a problem that didn’t exist.”
“What they’re trying to do is use that issue to create a permanent new infrastructure in Washington,” he added, calling it “the classic ‘never let a crisis go to waste’ strategy.”
The post In turf battle over AI, U.S. spy agencies vie for more sway than Commerce appeared first on Washington Post.
