Trump Signs Delayed AI Cybersecurity Order, Seeks Voluntary Access to Models

President Donald Trump signed an executive order on June 2, 2026, directing U.S. agencies to collaborate with artificial intelligence developers on cybersecurity, a directive he postponed in May over concerns it would hinder innovation. The order allows the government voluntary access to cutting-edge AI models for up to 30 days before public release to test for vulnerabilities. The policy establishes a clearinghouse for industry and agencies, including the Treasury and Pentagon, to coordinate software vulnerability responses. The National Security Agency will lead efforts to set standards for advanced models qualifying under the order, shortening the review period from an initial 90 days. Participation remains voluntary after earlier drafts considered mandatory government approval, reflecting pushback from tech leaders like former AI czar David Sacks. The move follows Anthropic’s revelation that its Mythos model discovered 6,202 critical vulnerabilities in open-source software. Mythos access will expand to 150 organizations. The Pentagon must collaborate with private-sector firms, though it remains in a dispute with Anthropic over contract guidelines. Companies including Alphabet Inc.’s Google, Microsoft Corp., and xAI already participate in Commerce Department’s voluntary AI evaluation program.