Anthropic embeds engineers at NSA to deploy Mythos AI for cyber operations

Anthropic, the company that built its brand on AI safety, now has approximately six engineers sitting inside the National Security Agency helping deploy an AI model designed for offensive cyber operations. The model is called Mythos, and it is already running on classified networks.

Here’s the thing: the Department of Defense, which oversees the NSA, classified Anthropic as a “supply chain risk” as of March 2026 and has been pursuing legal action against the company. The agency is simultaneously relying on Anthropic’s personnel to operate one of its most advanced cyber tools.

What Mythos actually does

Mythos was first introduced to a limited group in April 2026. It is an AI system purpose-built for finding and exploiting software vulnerabilities.

Early NSA testing has focused on identifying flaws in Microsoft products. Mythos can find security holes in widely used software faster than human analysts can.

The model has since been extended to 150 organizations across 15 countries. The NSA is running what’s called Mythos Preview on its classified networks, primarily for cybersecurity tasks.

The DoD paradox

Between March and April 2026, the DoD cited supply-chain risk issues with Anthropic and began pursuing legal action. Weeks later, the NSA started deploying Anthropic’s newest model with Anthropic engineers embedded on-site.

Why crypto and DeFi should pay attention

Industry analysts have noted that Mythos’s capabilities are forcing a reassessment of security practices across decentralized finance and blockchain systems. The shift is from smart-contract vulnerabilities toward infrastructure-level risks: key management systems, cross-chain bridges, and the underlying plumbing that connects blockchain ecosystems.

No specific cryptocurrencies or protocols have been directly linked to the Anthropic-NSA collaboration. But security researchers in the blockchain space are already recalibrating their threat models.

What investors should actually watch

On one side, the emergence of AI-powered offensive cyber tools creates new categories of risk for any protocol or platform that relies on software infrastructure. On the other side, if Mythos-class tools become widespread across 150 organizations in 15 countries, the demand for defensive capabilities that can match AI-speed attacks will grow proportionally.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.