President Donald Trump privately signed a landmark executive order on Tuesday establishing a voluntary federal testing framework for the world’s most powerful artificial intelligence (AI) systems before their public release.
The directive, which seeks to counter escalating cybersecurity threats posed by rapidly advancing AI technology, represents a compromise following weeks of internal administration debate. It arrives less than a month after the president abruptly canceled a high-profile signing ceremony for a more stringent draft of the order.
Under the finalized text, leading AI developers like OpenAI, Anthropic, and Google will be asked to submit their most advanced models to a voluntary government review 30 days prior to public deployment. This window is intended to give federal agencies, including the Pentagon and the National Security Agency (NSA), time to evaluate potential vulnerabilities to critical national security and financial infrastructure.
A previous draft slated for signature on May 21 required a 90-day pre-release review. Trump scrapped that version at the eleventh hour after warnings from tech executives and tech adviser David Sacks that a three-month delay would cripple American competitiveness against China.
“We’re leading China, we’re leading everybody, and I don’t want to do anything that’s going to get in the way of that lead,” Trump previously told reporters.
To appease industry concerns, the final text explicitly bars the government from establishing any “mandatory governmental licensing, preclearance, or permitting requirement.”
The administration’s urgency follows the April release of Anthropic’s Mythos Preview model, which alarmed Washington by demonstrating a superhuman ability to identify severe software vulnerabilities.
To mitigate these risks, the executive order mandates several coordinated federal actions. The Treasury Department and Department of Homeland Security will partner with private industry to centralize and patch software vulnerabilities within 30 days. NSA and the Cybersecurity and Infrastructure Security Agency (CISA) will oversee a classified process to identify which advanced frontier models fall under the scope of this policy. The Attorney General is directed to prioritize the prosecution of cybercrimes involving autonomous AI agents used to unlawfully access data.
White House spokesperson Liz Huston praised the directive, calling it a “common-sense approach… cementing America’s continued global dominance in AI.”
“I am highly encouraged by the directive for the DOJ to aggressively disrupt cybercriminals. Because Russia remains a safe haven for these syndicates, we must apply the full weight of U.S. government power to counter them,” said Tom Kellermann, vice president of AI security and threat research at TrendAI. “Furthermore, the new AI Clearinghouse is a positive initiative that will strongly support the work of CISA and the JCDC. Public-private partnerships and real-time threat sharing are the only ways to truly civilize cyberspace.”
External policy experts also offered support. Kevin Frazier, an AI law fellow at the University of Texas, noted that the order “aligns with a sensible, evidence-driven approach,” shifting focus away from partisan debates toward national resilience.
However, the order has drawn criticism from both sides of the regulatory debate. Some tech advocates remain skeptical of any federal encroachment, with former Trump AI adviser Dean Ball noting on X that the final text remains strikingly similar to the rejected May draft, questioning what the intelligence community could realistically accomplish in just 30 days.
Meanwhile, critics outside the administration worry the voluntary nature of the order signals that the U.S. is moving too slowly to contain the existential risks of unmonitored AI.
“This executive order reflects a growing reality: as AI accelerates both cyber defense and cyberattacks, organizations have less time to respond and must assume some threats will get through. The real challenge isn’t detection, it’s containment,” Gary Barlet, public sector chief technology officer at Illumio, said in an email.
Chimed in Dan Schiappa, president of technology and services at Arctic Wolf: “Regulation can be a double-edged sword. While governments have a role to play in promoting responsible AI development, we have to recognize that threat actors aren’t going to follow executive orders, regulations, or guardrails. The greater challenge is that adversaries are already using AI to increase the speed, scale and automation of cyberattacks, while many organizations are still relying on human-speed defenses.”
