Summary
The White House is expected to issue an executive order as early as today (Thursday, May 21) establishing a voluntary framework for government review of frontier AI models before public release. Under the proposed order, AI companies would share advanced models with federal agencies for testing dangerous capabilities and identifying vulnerabilities — with one draft specifying a 90-day pre-launch review period, though some companies are pushing for a shorter 14-day window.
The executive order is divided into two sections: a cybersecurity component that creates a voluntary “clearinghouse” managed by the Treasury Department and other agencies to find and fix security vulnerabilities in unreleased AI models, and a “covered frontier models” section that defines which AI systems fall under the review framework. The order also calls for expanded hiring at the US Tech Force, a body of engineers working to modernize government IT.
Major AI companies including OpenAI and Anthropic have been engaging directly with the White House on the order’s terms. The move follows Anthropic’s release of Claude Mythos — a model with unprecedented cybersecurity exploitation capabilities — and growing bipartisan concern about the offensive potential of frontier AI systems.
Source
Egypt Independent (via CNN) · CNN – Anthropic Mythos Context
Commentary
This is a notable shift for an administration that has been conspicuously hands-off on AI regulation. The catalyst is clear: Claude Mythos demonstrated that frontier models can identify and exploit cybersecurity vulnerabilities at a pace that fundamentally changes the threat landscape. When your AI can hack faster than your defenders can patch, the “let the market sort it out” approach starts looking untenable.
The voluntary nature is the key detail. Without enforcement teeth, this framework relies on goodwill from companies that have strong competitive incentives to ship fast. The 90-day vs. 14-day debate perfectly encapsulates the tension — the government wants meaningful review time, but companies argue that three months of delay in the current AI race is an eternity. Watch how this plays out with open-source and international labs that have no obligation to participate. A voluntary US framework means nothing if the next Mythos-equivalent ships from a lab in Shenzhen or Paris.