President Donald Trump signed an executive order establishing a voluntary pathway for federal agencies to test frontier artificial-intelligence models before they are released to the public, seeking to bolster cybersecurity without imposing formal licensing or preclearance regimes. The directive tasks the Pentagon, Treasury and CISA, among others, with creating a government-led evaluation program and an AI cybersecurity clearinghouse to share information on vulnerabilities and patches with industry. Companies would provide model access to the government up to 30 days ahead of broader distribution, a shorter window than an earlier draft’s 90 days.
The administration framed the move as pro-innovation and national-security minded, after delaying a planned May signing over concerns about hampering U.S. competitiveness versus China. The order also instructs the Justice Department to prioritize crimes involving AI, especially cyber intrusions and misuse of autonomous agents. The effort lands amid rapid advances—Anthropic’s Mythos Preview model has demonstrated superhuman vulnerability discovery—and expanding defensive partnerships across the private sector. While the White House courts closer collaboration with AI developers, it remains in a legal tussle with Anthropic over use restrictions tied to military and surveillance applications, underscoring the frictions between national-security aims and industry guardrails.
Related articles:
