Weeks after postponing a previous draft of the same order, President Donald Trump signed an executive order on June 2 asking artificial intelligence (AI) companies to voluntarily submit their frontier models for government review 30 days before a full public release.

The order asks some AI firms to participate in a voluntary evaluation process that aims to assess each model’s “advanced cyber capabilities” to determine at which point it could be designated as a “covered frontier model,” or one that could potentially pose cybersecurity or critical infrastructure risks.

Those firms whose models are evaluated to be “covered frontier models” will then be asked to collaborate with the federal government to select “trusted partners” that will gain access to the new models to “promote secure innovation and strengthen the cybersecurity of critical infrastructure” before the public gains access.

Here is a breakdown of the details contained in Trump’s order, some of the recent developments in the industry that inspired it, and how some lawmakers are reacting to the executive action. 

Order’s Details

If AI firms choose to participate in Trump’s governmental review, the process would entail “provid[ing] the Federal Government with access to covered frontier models, subject to appropriate confidentiality, cybersecurity, insider-risk, and intellectual-property protection, use, and nondisclosure requirements, for a period of up to 30 days before they plan to release such models to other trusted partners.”

Trump’s order also creates an “AI cybersecurity clearinghouse” that would function in “voluntary collaboration” with the AI industry and other critical infrastructure operators. 

The goal would be to scan for software vulnerabilities in frontier AI models while prioritizing “remediation and distribution of vulnerability patches.”

While the order designates various teams of executive branch agency heads to collaborate on this process, it’s unclear how exactly the models will be evaluated, what specific criteria designate one as a “covered frontier model” capable of advanced cybersecurity risks, and how the government will respond if any critical safety breaches are discovered.

Trump’s order makes it clear that nothing contained within “shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models.”

Trump had planned to sign a previous version of this executive order, but said on May 21 that he would delay the signing after telling reporters he had become dissatisfied with “certain aspects of it.”

AI Industry Developments

Focus and scrutiny on frontier AI models kicked into high gear when Anthropic announced its Claude Mythos Preview on April 7, the latest iteration of its Claude large language model (LLM). 

Not just pitched as the most capable version of Claude, Mythos is “strikingly capable at computer security tasks,” Anthropic said at the time—so much so that Mythos can exploit “zero-day vulnerabilities in every major operating system and every major web browser when directed by a user to do so,” the firm explained.

Zero-day exploits are vulnerabilities that are so difficult to detect that software developers are unaware of them at launch.

“Many of them are ten or twenty years old, with the oldest we have found so far being a now-patched 27-year-old bug in OpenBSD—an operating system known primarily for its security,” Anthropic said, describing some of the exploits as severe.

“We did not explicitly train Mythos Preview to have these capabilities. Rather, they emerged as a downstream consequence of general improvements in code, reasoning, and autonomy,” the firm added. “The same improvements that make the model substantially more effective at patching vulnerabilities also make it substantially more effective at exploiting them.”

The federal government directly acknowledged the risks almost a month later when the Commerce Department announced its Center for AI Standards and Innovation was partnering with Google DeepMind, Microsoft, and xAI to evaluate their frontier models for security risks ahead of public releases. 

The agency said the agreements build on partnerships that it made with OpenAI and Anthropic during the Biden administration in August 2024.

Kevin Kong, founder and CEO of Everstar Inc.—a technology firm that includes members from the Nuclear Regulatory Commission, CIA, Google, Microsoft, SpaceX, and Tesla—said that frontier AI is not just more powerful by “orders of magnitude,” it’s also more energy-intensive than previous generations.

“They’re also more agentic, as in they need to remember and forget the right information as it computes across time, and so this is why memory stocks have been going through the roof, because they need more and more memory chips, as well as the processing compute GPUs,” he told The Epoch Times.

More memory and processing power do not just translate into higher energy consumption, but also more of the infrastructure AI needs to operate and the money required to fund it. 

That’s likely why many AI firms are moving towards public valuation.

Anthropic said on Monday that it had confidentially filed an IPO prospectus with the Securities and Exchange Commission, meaning the company can go public once the agency completes its review.

OpenAI has indicated it may soon follow suit.

Reactions to Order

The Alliance for Secure AI was one of the first industry players to respond to Trump’s executive order on Tuesday.

The nonprofit, which “educates the public about the implications of advanced AI,” called on Congress to codify the order and make the government review process mandatory instead of voluntary.

Since Trump’s executive order allows AI companies to submit their frontier models to government review voluntarily, it’s not a compulsory process.

“After the national security wake-up call from advanced AI models like Mythos, we are pleased to see that the Trump administration is taking the risks of these models seriously. However, we know that Big Tech will still try to cut corners on safety and security,” Brendan Steinhauser, CEO of The Alliance for Secure AI, said in a statement.

“The next AI models will be even more powerful and will pose even bigger threats to our country than Mythos. These companies need oversight and cannot be trusted to do the right thing voluntarily.”

In a statement posted to X Tuesday afternoon, Sen. Bernie Sanders (I-Vt.) said it was “good news” that Trump had “finally acknowledged AI poses a real threat” after the president had previously condemned efforts to regulate the AI industry.

“The bad news? His executive order is voluntary and does almost nothing to protect Americans. Congress MUST act,” Sanders wrote.

California Gov. Gavin Newsom, considered by many to be a contender for the 2028 Democratic presidential primary, gave a similar reaction on X hours after Trump signed the order.

“After spending a year mocking AI safety concerns and calling them anti-American, the Trump Administration has finally conceded that guardrails are needed,” Newsom said.

“This EO is nowhere near California’s approach, but at least Trump is starting to acknowledge that AI shouldn’t regulate itself!”

David Sacks, who previously served as Trump’s AI and crypto czar before stepping down in March, praised the order for its limited regulatory oversight, noting that it “expressly forbids the creation of a new licensing, preclearance, or permitting regime.”

Kong, founder and CEO of Everstar, said on Tuesday that he had not yet read Trump’s full order, but said he thinks “it’s good that it’s a voluntary basis.”

“That’s how the Western world should operate, not like forcing private companies to do what the government thinks is its bidding,” he told The Epoch Times.

Kong said he thinks most foundational AI models will choose to participate.