Trump signs order giving NSA central role in voluntary pre-release reviews of advanced AI

President Donald Trump on Monday signed an executive order directing the federal government to expand AI-related cyber defenses and create a voluntary pre-release review process for the most advanced AI models, giving the National Security Agency a central role in deciding which systems warrant heightened scrutiny.

The June 2 order, titled “Promoting Advanced Artificial Intelligence Innovation and Security” and signed “DONALD J. TRUMP,” pairs a push for faster cybersecurity action across government with a new process for identifying especially powerful AI systems. The move matters for AI developers and federal agencies alike because it sets up a formal channel for companies to ask the government whether a model will be treated as a top-tier security concern, while stopping short of any mandatory licensing regime.

“The United States continues to lead the world in Artificial Intelligence (AI) because of the enormous talent and innovation of our AI industry, and because we refuse to stifle this innovation with overly burdensome regulation,” the order says.

The most consequential changes are in Section 3. Within 60 days, the Treasury secretary, the administration’s “Secretary of War” acting through the NSA director, and the Homeland Security secretary acting through the Cybersecurity and Infrastructure Security Agency must create a classified benchmarking process to assess the advanced cyber capabilities of AI models. That process is intended to set thresholds for designating a model a “covered frontier model” — effectively, the category of systems the government believes deserve closer review because of their cyber-related capabilities.

The order specifies that the NSA director will decide which models receive that designation, in consultation with other officials named in the order. It also directs the government to build a voluntary framework with AI developers so companies can ask in advance whether a model would qualify, provide government access to such models for up to 30 days before a planned release, and work with the government to identify trusted partners that could get early access for secure innovation and critical infrastructure protection.

The order says that early-access framework must include safeguards for confidentiality, cybersecurity, insider-risk and intellectual property. It also draws a bright line against turning the process into a gatekeeping system for model releases. “Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models,” Section 3(c) says.

Beyond model review, Section 2 sets a series of near-term deadlines for federal cyber defenses. Within 30 days, the Committee on National Security Systems must prioritize protection of national security systems, and the administration’s “Secretary of War” must do the same for that department’s own information systems. Within the same 30-day window, Homeland Security, through CISA, must issue Binding Operational Directives and related guidance to speed cyber defense of civilian federal systems and expand federal programs that use AI-enabled defensive tools.

That guidance is also meant to improve access to cybersecurity tools and services for federal, state and local governments and for critical infrastructure operators, including advanced AI models where appropriate. Also within 30 days, the Treasury Department, working with the NSA and CISA, must establish an “AI cybersecurity clearinghouse” with voluntary industry participation to coordinate vulnerability scanning, validation, remediation and patch distribution. Within 60 days, the Office of Personnel Management must expand hiring and placement pathways for federal cybersecurity specialists.

Section 4 addresses enforcement. It tells Attorney General Pam Bondi to prioritize use of existing federal criminal laws against people who use AI to illegally access or damage computers without authorization, or while committing related cybercrimes. The order cites existing statutes covering offenses such as identity and document fraud, computer fraud and wire fraud. It does not create new federal crimes.

The action follows May reporting that major AI companies, including Microsoft, Google and xAI, had agreed to provide early access to some new models to U.S. government reviewers on a voluntary basis for security evaluations. Trump’s order formalizes a broader federal framework around that idea.

It also comes after reported internal debate over whether advanced-model testing should sit mainly with civilian agencies such as the Commerce Department and its standards arm, the National Institute of Standards and Technology, or with the national security apparatus. The new order adopts a blended approach but gives the NSA a key role in deciding which models qualify as covered frontier models.

The order repeatedly uses “Secretary of War” and “Department of War,” terminology the administration authorized in 2025 as secondary titles in official correspondence and other non-statutory materials; Congress has not changed the department’s legal name from the Department of Defense.

That framing fits the administration’s broader AI posture: encourage development, add security measures and reject mandatory licensing. For companies weighing whether to cooperate, the clearest signal may be the order’s repeated insistence that the new review framework is voluntary and is not a permit system for releasing AI models.