The NSA is reportedly using Anthropic's Mythos - a restricted AI model not publicly available - despite an ongoing dispute between Anthropic and the Pentagon over the company's government contracts.
Mythos is separate from Anthropic's consumer-facing Claude models. Government-restricted AI systems like this typically require security clearances to access and operate under terms that differ significantly from commercial deployments. The public hasn't seen benchmarks, pricing, or even a feature description for Mythos - it exists entirely outside the normal product announcement cycle.
The timing makes this harder to dismiss as routine procurement. When a contractor and a government agency are in an active public dispute over contract terms, continued adoption by a separate intelligence agency - the NSA sits under the Department of Defense umbrella - suggests the tool's actual utility is outweighing the institutional friction. Intelligence agencies don't deploy new AI systems as a gesture of goodwill. If the NSA is actively using Mythos, someone decided it does something useful enough to push past the procurement complications.
Federal agencies have been moving quickly to put large language models (AI systems trained on vast text datasets to understand and generate human language) to work on intelligence analysis, document processing, and signals interpretation. The NSA, which handles enormous volumes of intercepted communications data, has obvious applications for these tools.
What the Pentagon feud is actually about remains unclear from what's been reported. A dispute that doesn't stop NSA adoption could mean the disagreement is about contract pricing or liability terms rather than the technology itself - a negotiating posture rather than a genuine break. Either way, the NSA's continued use is a meaningful data point for how government AI adoption works in practice: agency by agency, use case by use case, regardless of what's happening at the policy level above.