Trump Signs AI Executive Order on Cybersecurity and Model Review
In brief
- Trump signed an AI executive order Tuesday targeting cybersecurity and advanced model safety.
- NSA voluntary review process allows developers 30 days to submit advanced AI models for evaluation.
- Agencies will accelerate AI-powered cybersecurity tools and establish an AI clearinghouse.
- Critics argue the framework relies too heavily on voluntary cooperation from AI companies.
Voluntary Framework Over Mandatory Controls
The executive order seeks to reassure AI developers that the new framework will not create a formal approval process for releasing new models. Developers can voluntarily provide advanced AI models to the government for evaluation for up to 30 days before releasing them to other trusted partners. This approach reflects Trump's earlier concerns about regulatory overreach.
In May, Trump delayed signing a similar executive order, citing concerns that parts of the proposal could slow U.S. AI development and weaken America's position in competition with China. The latest version appears designed to address those worries by keeping the review process optional rather than mandatory.
Addressing Advanced Model Risks
The executive order establishes a classified review process under which the National Security Agency would determine whether advanced AI systems qualify as covered frontier models. The effort to draft an AI-focused executive order gained momentum after concerns surrounding Anthropic's Claude Mythos model, which demonstrated an ability to identify software vulnerabilities.
In April, U.S. Treasury Secretary Scott Bessent and then Federal Reserve Chair Jerome Powell reportedly convened a meeting with Wall Street bank CEOs to warn about cybersecurity risks tied to a new AI model. Anthropic has continued to roll out limited access to Mythos through Project Glasswing, a program meant to let tech and security firms and governments discover and address potential exploits. The company hinted that the public launch of Claude Mythos would occur in the coming weeks.
Enforcement and Criticism
The order calls for tougher enforcement against criminal uses of AI, including breaching information technology systems or employing AI agents to unlawfully access data for criminal purposes. Last month, federal prosecutors charged two men with using AI to generate and distribute sexually explicit images of women without their consent under the Take It Down Act.
Critics argue the framework relies too heavily on voluntary cooperation from AI companies. J.B. Branch, AI governance and technology policy counsel at Public Citizen, called for comprehensive federal AI legislation with enforceable safeguards and independent testing. The debate over whether voluntary measures or regulatory mandates better protect national security remains unsettled.
Frequently asked questions
Why did Trump delay this executive order in May?
Trump delayed the order citing concerns that parts of the proposal could slow U.S. AI development and weaken America's position in competition with China. The final version keeps the review process voluntary to address those worries.
How does the voluntary review process work?
Developers can voluntarily provide advanced AI models to the government for evaluation for up to 30 days before releasing them to other trusted partners. The National Security Agency determines whether models qualify as covered frontier models under a classified review process.
What prompted this AI-focused executive order?
Concerns surrounding Anthropic's Claude Mythos model, which demonstrated an ability to identify software vulnerabilities, gained momentum after Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened a meeting with Wall Street CEOs to warn about cybersecurity risks.
