President Donald Trump signed an executive order on June 2, 2026, establishing a security framework for artificial intelligence while specifically rejecting mandatory government licensing or pre-clearance requirements for new models. The order focuses on leveraging AI to strengthen federal and private computer systems, directing government secretaries to bolster cyber defenses. It also creates an AI cybersecurity clearinghouse for voluntary collaboration with industry and critical infrastructure operators to identify and mitigate vulnerabilities. This policy shift follows the April 2026 announcement by Anthropic that its latest Claude model, Mythos, autonomously identified hundreds of software vulnerabilities and simulated attacks across U.S. systems.

Under the new directive, the government will maintain a classified process for designating new programs as frontier models—AI systems trained on massive data sets that possess reasoning capabilities and autonomous tool-use. Developers of models meeting these criteria are directed to provide the government access at least 30 days before public release. Companies are also expected to collaborate with the federal government to engage third parties for risk assessment. However, the order maintains a voluntary reporting structure, continuing the administration’s stance that restrictive safety barriers might impede innovation. This contrasts with concerns from experts like Geoffrey Hinton and Yoshua Bengio, who argue that commercial pressures often prioritize speed over safety, potentially creating an evidence dilemma where slow policy implementation leaves societies vulnerable.

The executive order aligns with expert consensus by advocating for layered security, including hardware infrastructure tracking, rigorous pre-deployment testing, and real-time monitoring of systems. It acknowledges that advanced AI tools present a dual-use challenge, offering significant potential in fields like healthcare and defense while simultaneously providing new capabilities for malicious actors. Despite these provisions, the order does not address international cooperation, multilateral coordination, or globally shared governance, prioritizing instead the cultivation of domestic AI capabilities as a competitive national asset. The administration’s focus remains distinct from international processes like the AI Safety Summits held in the U.K., South Korea, France, and India, which involve a broader network of government-led safety institutes.

President Donald Trump signed an executive order on June 2, 2026, establishing a security framework for artificial intelligence while specifically rejecting mandatory government licensing or pre-clearance requirements for new models. The order focuses on leveraging AI to strengthen federal and private computer systems, directing government secretaries to bolster cyber defenses. It also creates an AI cybersecurity clearinghouse for voluntary collaboration with industry and critical infrastructure operators to identify and mitigate vulnerabilities. This policy shift follows the April 2026 announcement by Anthropic that its latest Claude model, Mythos, autonomously identified hundreds of software vulnerabilities and simulated attacks across U.S. systems.

Under the new directive, the government will maintain a classified process for designating new programs as frontier models—AI systems trained on massive data sets that possess reasoning capabilities and autonomous tool-use. Developers of models meeting these criteria are directed to provide the government access at least 30 days before public release. Companies are also expected to collaborate with the federal government to engage third parties for risk assessment. However, the order maintains a voluntary reporting structure, continuing the administration’s stance that restrictive safety barriers might impede innovation. This contrasts with concerns from experts like Geoffrey Hinton and Yoshua Bengio, who argue that commercial pressures often prioritize speed over safety, potentially creating an evidence dilemma where slow policy implementation leaves societies vulnerable.

The executive order aligns with expert consensus by advocating for layered security, including hardware infrastructure tracking, rigorous pre-deployment testing, and real-time monitoring of systems. It acknowledges that advanced AI tools present a dual-use challenge, offering significant potential in fields like healthcare and defense while simultaneously providing new capabilities for malicious actors. Despite these provisions, the order does not address international cooperation, multilateral coordination, or globally shared governance, prioritizing instead the cultivation of domestic AI capabilities as a competitive national asset. The administration’s focus remains distinct from international processes like the AI Safety Summits held in the U.K., South Korea, France, and India, which involve a broader network of government-led safety institutes.