White House Considers Pre-Release Vetting of AI Models in Sharp Policy Reversal

--- headline: "White House Considers Pre-Release Vetting of AI Models in Sharp Policy Reversal" slug: white-house-ai-model-vetting-pre-release category: business story_number: "01" date: 2026-05-04 ---

The Trump administration is discussing an executive order that would require U.S. government review of advanced artificial intelligence models before they are released to the public, according to a New York Times report published Sunday -- a dramatic about-face from the president who revoked his predecessor's AI safety requirements on his first day back in office sixteen months ago.

The Proposal

Senior White House officials briefed executives from Anthropic, Alphabet's Google, and OpenAI on the plans during meetings last week, according to U.S. officials and people briefed on the matter. The proposed executive order would establish a working group of tech executives and government officials tasked with creating a standardized "red-teaming" process in which federal experts audit a model's capabilities before it is ever launched publicly.

The agencies under consideration to oversee these reviews include the National Security Agency, the Office of the National Cyber Director, and the Office of the Director of National Intelligence. The proposed system could grant government agencies early access to models without necessarily blocking releases outright -- a structure that more closely resembles the United Kingdom's multi-layered AI oversight model, where the AI Safety Institute serves as a supplementary review body rather than a hard gatekeeper.

A White House official dismissed the reporting as premature, saying that any policy announcement "will come directly from the president" and that discussion about "potential executive orders is speculation."

The Mythos Catalyst

The policy reversal did not emerge in a vacuum. It follows weeks of escalating concern over Anthropic's Mythos model, which cybersecurity researchers have called the most capable AI system ever built for finding and exploiting software vulnerabilities. Since its limited preview release in early April under Project Glasswing -- a $100 million defensive cybersecurity initiative backed by Anthropic -- Mythos Preview has discovered thousands of high-severity zero-day vulnerabilities across every major operating system and web browser, including a 27-year-old bug in OpenBSD and a 16-year-old flaw in FFmpeg.

Anthropic CEO Dario Amodei has positioned the company's cautious rollout as a model for responsible deployment. "The dangers of getting this wrong are obvious, but if we get it right, there is a real opportunity to create a fundamentally more secure internet and world than we had before the advent of AI-powered cyber capabilities," Amodei wrote on X when announcing Project Glasswing. In a separate statement, he warned: "More powerful models are going to come from us and from others, and so we do need a plan to respond to this."

The White House's Office of the National Cyber Director hosted two meetings last week -- one with tech and cybersecurity companies, another with trade groups -- to discuss the security implications of models like Mythos. Amodei himself visited the White House in April for what officials described as a "productive and constructive" meeting with Chief of Staff Susie Wiles, and he has publicly called for AI to be regulated "the way you regulate cars and aeroplanes" through mandatory, third-party-assessed safety standards.

Why This Matters

The proposed policy would represent an extraordinary reversal of the administration's core AI philosophy. On January 20, 2025, Trump revoked Executive Order 14110 -- the Biden-era directive signed in October 2023 that required developers of AI systems posing national security risks to share safety test results with the government under the Defense Production Act before public release. Three days later, Trump signed his own executive order, titled "Removing Barriers to American Leadership in Artificial Intelligence," which emphasized innovation over regulation.

As recently as March 2026, the White House's National Policy Framework for Artificial Intelligence expressly recommended against creating any new federal rulemaking body to regulate AI, instead calling for governance through existing regulatory agencies and industry-led standards. A pre-release vetting regime would contradict that framework on its face.

Yet the Mythos episode appears to have forced a recalculation. The model's ability to autonomously discover and weaponize zero-day exploits -- capabilities that Anthropic itself called a "step change" beyond prior systems -- presented the administration with a concrete national security scenario that its deregulatory stance was not designed to address. The White House had already been scrambling since April to understand Mythos's hacking capabilities, even as it simultaneously tried to navigate the awkward reality that federal agencies were barred from using Anthropic's technology over a separate supply chain risk dispute.

The international dimension adds further pressure. The UK's AI Safety Institute has been conducting pre-release evaluations of frontier models from major labs since 2024, positioning Britain as a leader in AI governance. The European Union's AI Act, now in phased enforcement, imposes its own transparency and risk-management obligations on general-purpose AI models. If the U.S. adopts a pre-release review mechanism, it would bring American policy closer to its allies' frameworks after more than a year of deliberate divergence.

What to Watch Next

The key variable is whether the executive order materializes -- and in what form. A soft version that merely establishes a working group and voluntary review processes would preserve the administration's pro-innovation posture while creating political cover on safety. A harder version with mandatory pre-release audits by intelligence agencies would fundamentally reshape the relationship between government and AI labs, potentially slowing deployment timelines and raising questions about classified review processes and intellectual property protections.

Watch for signals from the three companies briefed on the plans. OpenAI, Google, and Anthropic each have different incentives: Anthropic has already endorsed the principle of government oversight, while OpenAI and Google may be warier of a process that could advantage incumbents by raising compliance barriers for smaller competitors. Also worth monitoring: whether Congress attempts to codify any executive order into legislation, and how the proposed review framework interacts with the administration's ongoing effort to resolve Anthropic's federal supply chain blacklist status. The intersection of those two tracks -- vetting new models while simultaneously trying to onboard the very company whose model triggered the vetting debate -- will reveal whether this is a durable policy shift or a reactive gesture to an unprecedented moment in AI capabilities.