Source: NBC News

The executive order signed in private instructs federal agencies including the Defense Department, the Treasury and CISA to strengthen cybersecurity for critical infrastructure and establishes a voluntary mechanism for the federal government to test and vet the most powerful AI systems for safety issues before deployment. The testing framework relies on voluntary collaboration from leading U.S. AI firms such as Anthropic, OpenAI and Google and explicitly bars the government from imposing a mandatory licensing or preclearance regime.

The order directs a coalition of administration offices spanning the Treasury, the NSA and the Department of Homeland Security to create an AI cybersecurity clearinghouse with industry collaboration to centralize knowledge of software vulnerabilities and patches. It sets a voluntary access window of up to 30 days before companies plan to release cutting‑edge models to other trusted partners (the original draft had allowed up to 90 days), directs the attorney general to prioritize prosecuting AI-related crimes with a focus on cybercrimes and use of autonomous AI agents for unlawful access, and notes that plans for the signing originally scheduled for May 21 were canceled before the private signing. The dossier reports that Anthropic’s Mythos Preview in April prompted concern in Washington for its ability to find critical vulnerabilities, that Anthropic expanded Project Glasswing access to 150 organizations across more than 15 countries, and that a court ruled in Anthropic’s favor in California while related litigation remains pending in Washington, D.C.