NSA Spent Billions on AI Model Security Testing

The agency prioritized identifying vulnerabilities in frontier AI models throughout 2026.

Updated on Sept. 24, 2026 in Artificial Intelligence

Isometric editorial illustration of a modular server rack array in a secure facility, representing national security infrastructure. — The National Security Agency spent billions throughout 2026 to stress-test frontier artificial intelligence models for national security vulnerabilities. AI Illustration. Upload story photo >

Live Poll

Should private AI companies, rather than taxpayers, pay the costs for federal AI safety testing?

Throughout 2026, the National Security Agency invested billions of dollars to test advanced artificial intelligence models for national security vulnerabilities. These efforts, conducted through the agency's Artificial Intelligence Security Center, relied on classified funding from the federal national security budget.

Why it matters

The agency aims to mitigate potential national security risks posed by frontier models as these systems grow more advanced. Testing these technologies has become a significant financial undertaking, with costs scaling alongside the computing power required for analysis.

The NSA's primary expenses for this initiative include purchasing extensive computing power and hiring specialized personnel. These expenditures are significant compared to proposed legislative costs, such as the $20 million annual estimate for the AI Security and Innovation Act.

The players

National Security Agency

An intelligence agency of the United States government responsible for global monitoring, collection, and processing of information for foreign and domestic intelligence and counterintelligence purposes.

Department of Defense

The federal executive department responsible for providing the military forces needed to deter war and ensure national security, managing an annual budget of approximately $1 trillion.

The details

The NSA conducts these assessments through its Artificial Intelligence Security Center, a facility dedicated to identifying vulnerabilities in frontier models—large-scale AI systems that exceed current state-of-the-art benchmarks. The agency utilizes classified segments of the federal budget to acquire the necessary infrastructure to stress-test these models. By simulating adversarial threats, the center monitors for potential system failures or security risks that could be exploited by hostile actors.

2026: The NSA spent billions of dollars on artificial intelligence model testing.
September 24, 2026: A report was published detailing classified intelligence estimates regarding these expenditures.

The Tech Race

The NSA's current multibillion-dollar testing initiative far exceeds the $20 million annual budget projected under the AI Security and Innovation Act. This discrepancy highlights the scale at which the federal government is competing to maintain dominance in frontier AI security.

While these testing programs are classified, the massive federal spend indicates a tightening regulatory and security environment for commercial AI developers. The high cost of verification may eventually influence how quickly private labs can deploy new models in government-adjacent sectors.

The takeaway

The sheer scale of federal AI spending suggests that security verification is becoming a dominant cost driver in the industry. Watch for future legislative debates regarding whether these billions in costs will be shifted to private labs through mandatory compliance reporting.