The National Security Agency (NSA) is a US intelligence agency responsible for signals intelligence (SIGINT) and information assurance / cybersecurity for federal government systems. It is a component of the Department of Defense and reports through the Director of National Intelligence for intelligence-community matters.
AI cybersecurity capability deployment
Per the April 2026 Amodei entry, "NSA is independently testing Mythos" for vulnerability discovery in Microsoft products, a frontier-AI cyber-capability use case operating in parallel with the contested public Anthropic-Pentagon dispute. The independent Mythos testing, dated to April 2026, targets vulnerability discovery in Microsoft products and runs alongside the public Pentagon-Anthropic procurement dispute. The testing target is Claude Mythos Preview.
AI Security Center (AISC)
The AI Security Center (AISC) is NSA's component focused on AI in national-security applications. It provides classified-AI guidance to the defense industrial base.
According to two sources familiar with classified intelligence estimates, the NSA has told lawmakers that it is spending billions of dollars this year evaluating and testing advanced AI models through the center; the disclosure became public on September 24, 2026. The same report compared the figure with Congressional Budget Office estimates of roughly $20 million per year for the AI Security and Innovation Act and $36 million in total over five years for a separate House AI reporting and tracking bill. The Defense Department declined to comment on NSA spending (Source: washingtonsun.com).
Congressional authorization activity
On or before July 24, 2026, the House Intelligence Committee passed an Intelligence Community authorization bill that would expand the NSA's AI Security Center and create a senior post to coordinate the agency's AI efforts (Source: insideaipolicy.com).
Reps. Josh Gottheimer and Mike Lawler announced on September 18, 2026 the American AI Security Act, which would require developers of the most powerful AI models to give the NSA full access for a mandatory pre-deployment review of whether a model could be used to launch a serious cyberattack or help build a chemical, biological or radiological weapon, within 30 days plus one 30-day extension, with an expedited appeal modeled on national-security courts (Source: gottheimer.house.gov). Gottheimer, a member of the House Intelligence Committee, argued that third-party testers lack national-security clearances.
Frontier-AI export-control input
NSA's SIGINT analysis informs BIS chip-smuggling enforcement and export-control rulemaking.
Joint advisory on adversarial distillation
On September 8, 2026 the NSA issued joint cybersecurity advisory AA26-251A with CISA and the FBI, attributing industrial-scale knowledge-distillation campaigns against US frontier models to six named China-based companies — DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI — since at least late 2024, "likely with Chinese government awareness." The advisory maps each firm to the US models and capabilities it distilled, maps eight techniques to MITRE ATLAS and describes four further techniques as absent from it, and recommends behavioral detection, undisclosed response degradation for confirmed distillation traffic, and cross-organization intelligence sharing (Joint Cybersecurity Advisory AA26-251A: China-Based AI Companies Conducting Industrial-Scale Distillation Campaigns (NSA/CISA/FBI, September 2026)). The signed PDF is hosted on media.defense.gov. See Adversarial Distillation.
Relationships
- related: Cybersecurity and Infrastructure Security Agency (CISA) — regulator role (civilian-cybersecurity counterpart), Chief Digital and AI Office (CDAO) (DOD-procurement counterpart), UK AI Safety Institute (AI Security Institute) (UK Five-Eyes peer).
- related: AI and Cybersecurity, AI and National Security, Claude Mythos Preview (NSA testing target).
- related: DOD — Department of Defense (AI Deployer) (parent — for operational matters).
- related: Clawed (Anthropic-Pentagon dispute context).
Sources
Stub created 2026-05-11. Foundational ingest candidates: NSA AISC public guidance, NIST AI cybersecurity standards co-authored by NSA, Five Eyes-coordinated AI publications.