U.S. Administration Discusses Expanding Intelligence Agencies' Role in AI Evaluations

By Cat Zakrzewski

*Published: 2026-05-11T17:54:31Z*

*Updated: 2026-05-13*

As President Donald Trump prepared for a summit with Chinese President Xi Jinping, administration officials discussed a proposal to expand the role of U.S. intelligence agencies in evaluating advanced AI models, according to two people familiar with the internal discussions who spoke on the condition of anonymity because the proposal was not public at the time.

Trump considered an executive order on AI security, with discussions pointing to a potential signing as early as May 12, 2026. However, no such executive order had been signed as of May 13, according to White House records and public announcements. The administration's approach to advanced AI models remained under discussion among officials, according to multiple people with knowledge of the conversations.

The discussions involved officials from the Commerce Department and national security aides, who differed on which agency should lead evaluations of AI models. Silicon Valley executives have described such models as capable of transforming the economy.

The talks gained attention ahead of Trump's meeting with Xi and as U.S. companies anticipated releases of advanced models from Chinese competitors, comparable to Anthropic’s Mythos and others.

Administration officials recently addressed cybersecurity risks from Mythos and similar models. Anthropic stated that Mythos outperforms humans in identifying software vulnerabilities, raising questions about its potential use by hackers targeting critical infrastructure such as the energy grid, banks, and government systems.

In response, the White House Office of the National Cyber Director proposed establishing a center within the Office of the Director of National Intelligence (ODNI) to evaluate new AI models, which would increase intelligence agencies' involvement in AI assessments.

Officials at the Commerce Department, which operates the Center for AI Standards and Innovation (CAISI) for evaluating models through voluntary company agreements, raised objections to the proposal, according to some of the people familiar with the discussions 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. “From export controls to the AI exports program, we are laser focused on ensuring American leadership.” The ODNI referred inquiries to the White House.

CAISI has developed an evaluation system and hired AI experts with advanced degrees, according to one person familiar with the agency. Commerce officials have cited this infrastructure as positioning the department to handle testing. The intelligence community maintains expertise in cybersecurity and AI through its technical personnel.

The Commerce Department, particularly under the Trump administration, has emphasized policies supporting industry competitiveness against China. Transferring evaluations to intelligence agencies could result in additional focus on cybersecurity and national security risks of AI models.

Anthropic’s Mythos announcement prompted administration reviews of its prior approach to AI oversight, which involved rescinding several measures from President Joe Biden’s administration.

“They’re relitigating everything on AI policy right now,” said Chris McGuire, senior fellow at the Council on Foreign Relations and a former 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?”

A White House official, speaking anonymously about private deliberations, said the administration is consulting across government and industry, including frontier AI labs. “Any policy announcement will come directly from the president,” the official said. “Discussion about potential executive orders is speculation.”

On May 5, 2026, CAISI publicly announced partnerships with Google DeepMind, Microsoft, and xAI to conduct pre-deployment evaluations and research on frontier AI capabilities and security. The agency’s website briefly displayed an error page stating, “Oops, that’s not standard?! The page you requested cannot be found at this time. It may be temporarily unavailable or it may have been removed or relocated.” As of May 11, the page remained inaccessible, according to one person familiar with the matter. However, the partnerships proceeded without reported disruptions, and CAISI continued its evaluation activities.

Under Biden’s AI executive order, developers of advanced models were required to notify the government and share testing results with Commerce. The order established the U.S. AI Safety Institute under Commerce’s National Institute of Standards and Technology (NIST), part of an international network evaluating AI models. The Trump administration renamed it the Center for AI Standards and Innovation, removing “Safety” from the title.

Administration officials have weighed whether AI model evaluations should be mandatory or remain voluntary, as with current CAISI arrangements. “That’s very much a topic of active debate,” said one person familiar with the discussions. Some in the intelligence community and other areas advocate for mandatory evaluations.

Details on mandatory evaluations remain undecided, including whether to pursue legal authority or apply other forms of government leverage, the person said.

White House officials recently issued varying public statements on AI policy.

National Economic Council Director Kevin Hassett suggested evaluating AI models similarly to the Food and Drug Administration’s process for new drugs, ensuring safety before public release.

Hassett’s remarks drew industry reactions, as they contrasted with the administration’s lighter-touch regulatory stance on AI.

Hours later, White House Chief of Staff Susie Wiles posted on X, stating the president would not be “in the business of picking winners and losers.”

Trump allies in Silicon Valley interpreted Wiles’s statement as opposing an FDA-like model for AI. David Sacks, former White House AI lead who continues advising on AI policy, said on the “All-In” podcast that no senior officials support such a framework.

Sacks advocated deploying advanced models to cybersecurity firms to aid preparation for releases like Mythos later in 2026. He argued companies have acted responsibly, citing Anthropic and OpenAI. “No one was trying to release these super powerful models,” Sacks said. He opposed pre-release government approvals, stating companies are managing risks appropriately and that calls for new infrastructure resemble exploiting events for expanded regulation.

The Commerce Department’s partnerships with Google DeepMind, Microsoft, and xAI underscore its ongoing role in AI evaluations. These collaborations aim to assess capabilities and advance AI security standards, aligning with the administration’s focus on U.S. leadership.

Public statements from Commerce affirm alignment with Trump’s priorities, including export controls. Meanwhile, intelligence agencies’ proposed center would leverage their cybersecurity expertise amid concerns over models like Mythos.

As of May 13, no executive order on AI security has been issued, and discussions continue without resolution on mandatory testing or agency leads. The administration’s engagement with industry persists, reflecting broader efforts to balance innovation, security, and competition with China.

Trump’s summit preparations highlighted AI’s strategic importance. U.S. firms monitor Chinese developments, while domestic debates address how to evaluate models without hindering competitiveness.

CAISI’s infrastructure, built on NIST foundations, supports voluntary testing. Intelligence proponents emphasize risks requiring specialized oversight.

Industry figures like Sacks stress self-regulation, noting no prior incidents necessitated approvals. McGuire’s observations point to open questions on testing modalities.

White House clarifications via Wiles and officials indicate policy formation remains fluid, centered on presidential direction.

These dynamics occur against Biden-era changes, now under review post-Mythos. Commerce’s active partnerships demonstrate continuity in evaluations despite internal discussions.

(Word count: 1237)