How the Quiet Signing of New White House Exec Order Impacts AI Frontier Models
Keep the AI Lead, Says Bessent
By Rebecca Grant. President Donald J. Trump inked the new AI Executive Order on June 2. The EO recognizes that advanced AI can exploit cyber vulnerabilities and energizes the government, but leaves industry unshackled and in a strong position to help. The President “found an order he could live with,” judged Politico, also reporting on the “quiet” signing of the EO.
“Advanced AI capabilities make our Nation stronger, but also introduce new national security considerations that require coordinated action across executive departments and agencies (agencies), and components. As these capabilities evolve, my Administration will continue to work closely with industry to ensure that the best and most secure technology is deployed rapidly to confront any and all threats to our country,” said the EO.
The big takeaway is a voluntary framework or “clearinghouse”, where developers can consult with the government 30 days before they release frontier AI models. That’s down from a 90-day review in a previous draft. Powerful frontier models can penetrate weak spots in cybersecurity and software that is deployed across government departments. Industry was well aware of this problem. Project Glasswing was set up as an endeavor by 50 partners to secure critical software before AI can find and exploit its vulnerabilities. Companies are using Mythos Preview internally to find and fix their own bugs. “Advanced AI models are surfacing additional issues in code,” as Microsoft recently noted.
“I continue to think this EO is a mistake. This is clearly teeing up the infrastructure for a model licensing regime,” Dean W. Ball wrote on X. Ball was the senior drafter for the July 2025 White House AI Action Plan.
While the new EO was pending, current and former White House officials worked to clarify that government can’t get in the way of industry. Former White House AI adviser David Sacks has said pre-release government approvals were “solving a problem that didn’t really exist,” since Anthropic and other AI companies were already keeping Mythos-like models away from the public.
Secretary of the Treasury Scott Bessent boiled it down at a May 28 press conference. “The U.S. is the AI leader in the world. We are an AI superpower. China is second. They are trailing substantially. We want to make sure we keep that lead. So, we are working on the exact calculus between innovation and safety, and we want to optimize for that,” Bessent said. Bessent’s comments suggest the White House is taking a favorable view of Silicon Valley’s efforts to police up its own software vulnerabilities. “We’re working very closely with the large language labs; they’ve been excellent partners,” Bessent said.
Staying on top of vulnerabilities will take all hands on deck. For example, another issue surfacing is how to construct the “red team” threats. Cisco warned that single prompt attacks aren’t realistic because real-world attacks start with small inquiries then escalate over time.
Under this scenario, multi-turn jailbreaking attacks bypass AI security measures by gradually steering conversations toward harmful outputs through innocent-seeming steps, creating serious business risks that standard single-message testing misses.
As signed, the AI Executive Order acknowledges how difficult it is for traditional government regulation approaches – however well-intentioned – to cope with the innovation onslaught of AI. Today it’s language models; tomorrow, engineering workflow. This is just the first frontier of AI. America’s government must be careful to find a way to champion AI standards, but not by impounding innovation pending bureaucratic review.