NSA Using Anthropic's Cybersecurity Model for Offensive Operations, Report Says
The National Security Agency is using Anthropic's Mythos model for offensive cyber operations with help from roughly half a dozen Anthropic engineers embedded inside the agency, according to a Financial Times report citing people familiar with the arrangement.
The development marks a striking reversal. In March 2026, the Trump administration labeled Anthropic a "supply chain risk" and barred the company from selling to any branch of the Department of Defense. The company is currently suing the DOD over that decision.
The NSA's reported use of Mythos suggests either a shift in government posture toward Anthropic or disagreement within defense agencies about the model's value. Two courts have issued contradictory rulings on separate Anthropic-DOD lawsuits, further muddying the situation.
Why the NSA Wants Mythos
The rationale is straightforward: U.S. adversaries will use cybersecurity-focused AI models. Gaining early access provides tactical advantage in a threat environment where zero-day exploits are increasingly common.
Anthropic's embedded engineers are customizing Mythos for "specialized applications"-likely tailoring the model to specific types of cyber-attacks targeting nations such as China or Iran, according to the report. The sources were unclear whether the engineers are supporting active operations or simply guiding technology use and customization.
The Contract Dispute That Led Here
The conflict began in January 2026 when the DOD and Anthropic negotiated a $200 million contract. The Trump administration demanded Anthropic remove AI guardrails to allow use "for all lawful purposes."
CEO Dario Amodei refused, stating the company would not remove safeguards against mass surveillance and autonomous weapons. Amodei said that "in a narrow set of cases, AI can undermine, rather than defend, democratic values."
The DOD canceled the contract and moved to OpenAI instead. After public backlash, OpenAI added safety clauses to its deal-the same restrictions the DOD had rejected from Anthropic. This suggests the dispute may have been politically motivated rather than rooted in technical concerns.
What Operations Teams Should Know
The NSA arrangement shows how government agencies deploy AI models in ways that differ from commercial use. Mythos is currently in early access, distributed to 150 organizations across 15 countries after its April release.
Separately, the Trump administration issued an executive order requiring AI companies to share leading-edge models with the government for 30 days before public release. The order leaves undefined what qualifies as a "covered frontier model" and what happens if companies decline.
For operations professionals, this signals that AI governance and security implications remain unsettled across government. The contradictory court rulings and shifting agency positions create uncertainty about how AI tools will be regulated and deployed in critical infrastructure.
Understanding how models like Claude AI Courses are being used in government operations-and the policy tensions surrounding them-matters for anyone managing technology deployments. AI for Operations Managers covers these governance and strategic considerations directly.
Your membership also unlocks:
