The US government is moving to make artificial intelligence a more coordinated part of national security. A White House memorandum lays out how agencies should use AI, how they should evaluate risk, and how they should work together as the technology becomes more important to defense, intelligence and security systems.
The document is not only about adopting new tools. It also tries to set rules for how AI is bought, tested, staffed and governed across agencies, while keeping US leadership in the technology intact.
A national security AI strategy built around control and speed
The memorandum rests on three main goals. The first is to protect the US position in AI development. The second is to make AI useful for national security work. The third is to help shape an international framework for responsible AI development.
That combination shows the central tension in the strategy. The US government wants agencies to move faster, but it also wants a clearer structure for safety, ethics, acquisition and oversight. In practice, the memorandum treats AI as both a strategic advantage and a technology that requires disciplined governance.
Jake Sullivan, US National Security Advisor, framed the issue as a race to use existing advantages before they fade. In his speech at the National Defense University, he warned: "If we don’t act more intentionally to seize our advantages, if we don’t deploy AI more quickly and more comprehensively to strengthen our national security, we risk squandering our hard-earned lead."
The memorandum links that lead to several practical inputs. It points to attracting global AI talent, developing advanced chips and building the infrastructure needed for AI systems. Sullivan put the talent point directly: "America has to continue to be a magnet for global, scientific, and tech talent."
What agencies are being asked to change
The plan calls for major coordination across national security agencies. A key piece is the creation of an AI National Security Coordination Group with Chief AI Officers from major agencies. Its role is to harmonize policies for AI development, accreditation, acquisition, use and evaluation across national security systems.
The Department of Defense and the Office of the Director of National Intelligence are also directed to establish a working group on AI procurement issues for national security systems. That group is expected to develop recommendations for acquisition processes and later provide recommendations to the Federal Acquisition Regulatory Council.
The memorandum also addresses the human side of AI capacity. The Assistant to the President for National Security Affairs is tasked with streamlining visa processing for AI talent in sensitive technologies. The Coordination Group is directed to establish a Talent Committee to standardize hiring and retention of AI talent for national security.
Other agencies receive workforce responsibilities as well. The State Department and agencies are directed to identify AI training opportunities to increase workforce AI competencies. The Council of Economic Advisers is asked to analyze the domestic and international AI talent market, while the National Economic Council is tasked with assessing US private sector AI competitive advantages and risks.
For national security agencies, the message is clear: AI adoption is not just a technical deployment issue. It touches hiring, procurement, training, accreditation, evaluation and interoperability.
Safety testing becomes a central pillar
A major part of the memorandum focuses on testing AI systems for security risks. The AI Safety Institute is assigned a central role in evaluating frontier AI models, issuing guidance for testing and managing dual-use AI model risks, and developing benchmarks for capabilities relevant to national security.
The National Security Agency is directed to develop capability for testing AI models' potential for cyber threats. The Department of Energy and National Nuclear Security Administration are directed to develop testing capability for AI nuclear/radiological risks, including classified and unclassified tests.
The Department of Energy is also tasked with evaluating radiological/nuclear implications of frontier AI models and submitting assessments on AI nuclear/radiological risks and recommended actions. The Department of Energy, Department of Homeland Security and AI Safety Institute are directed to develop a roadmap for evaluating AI models' potential for chemical/biological threats.
The memorandum also includes work on biosafety and biosecurity. Multiple agencies, including the Department of Defense, Health and Human Services, Department of Energy, Department of Homeland Security and National Science Foundation, are directed to support AI efforts enhancing biosafety and biosecurity through various technical initiatives.
The Office of Science and Technology Policy, National Security Council staff and Office of Pandemic Preparedness are assigned guidance for safe in silico biological and chemical research. The National Science Foundation is also tasked with developing best practices for publishing potentially sensitive AI research.
Sullivan used the history of railroads to explain the role of standards in speeding adoption. He said: "During the early days of the railroads, for example, the establishment of safety standards enabled trains to run faster thanks to increased certainty, confidence, and compatibility."
The timetable turns broad goals into deadlines
The memorandum is detailed about timing. Within 30 Days, the Department of Defense and Office of the Director of National Intelligence are to establish the working group on AI procurement issues. Within 45 Days, the AI National Security Coordination Group is to be formed.
Within 90 Days, the plan calls for visa processing work for AI talent, a review of the Intelligence Priorities Framework to better assess foreign threats to the US AI ecosystem, and creation of the Talent Committee. By 120 Days, agencies are expected to identify AI training opportunities, while the National Security Agency and Department of Energy/National Nuclear Security Administration are expected to develop testing capabilities for cyber and nuclear/radiological risks.
The 180 Days mark is especially dense. It includes the Council of Economic Advisers talent market analysis, the National Economic Council assessment of private sector AI advantages and risks, a Department of Energy pilot project for federated AI training and inference, AI Safety Institute testing of frontier AI models, dual-use guidance, national security benchmarks and department-level AI governance guidance.
Other deadlines extend to 210 Days, 240 Days, 270 Days, 365 Days and 540 Days. The memorandum also creates ongoing and annual requirements, including annual updates to AI guidance, annual AI safety assessment reports, annual nuclear/radiological risk assessments, and annual reports for 5 years on AI integration and agency AI activities.
Allies, China and the global AI framework
The strategy is not limited to domestic agencies. The memorandum says the US wants to work with allies to create standards for secure and trustworthy AI development. The State Department is directed to create a strategy for international AI governance norms promoting safety and democratic values.
The Department of Defense is also asked to evaluate the feasibility of AI co-development with allies, including potential partners and testing vehicles. At the same time, the memorandum draws boundaries around technology transfer, especially in relation to China.
Sullivan described the balance this way: "We need to balance protecting cutting-edge AI technologies on the one hand, while also promoting AI technology adoption around the world," adding, "Protect and promote. We can and must and are doing both."
The result is a national security AI strategy that tries to do several things at once: accelerate deployment, protect sensitive capabilities, improve testing, coordinate agencies, attract talent and build international norms. Its success will depend on whether agencies can turn the timetable into working governance, not just new reporting lines.