Despite the Pentagon's high-profile designation of Anthropic as a "supply chain risk" in early 2026 — triggered by the company's refusal to remove ethical guardrails limiting military use for autonomous weapons and domestic surveillance — the National Security Agency has been actively using the firm's most advanced model, Mythos Preview. According to multiple reports, the NSA is among roughly 40 select organizations granted access to this restricted model, which excels at identifying exploitable security vulnerabilities and possesses significant offensive cyber capabilities that Anthropic has deliberately limited from broader release. This development directly contradicts the Department of Defense's earlier efforts to blacklist Anthropic, including directives that halted broader federal use of its Claude models.

The conflict escalated when Anthropic sued the Trump administration, arguing the supply-chain risk label was an unlawful retaliatory measure violating First Amendment protections and exceeding statutory authority under 10 USC 3252. Federal courts issued temporary stays but later denied some motions to fully lift the designation, leaving DoD contractors restricted in classified or offensive contexts even as other agencies advanced. Meanwhile, the White House Office of Management and Budget has directed federal agencies to prepare for Mythos deployment under controlled conditions, and Anthropic CEO Dario Amodei held a "productive" meeting with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent to discuss government integration, security practices, and the model's deployment. Treasury and Federal Reserve officials have expressed alarm that Mythos could "hack every major system," prompting urgent red-teaming and outreach to major banks about existential AI-powered threats to financial infrastructure.

UK counterparts have similarly secured access via the British AI Security Institute, underscoring allied interest in these dual-use capabilities. What mainstream coverage frames as bureaucratic inconsistency reveals deeper dynamics: the U.S. government's increasing dependence on a handful of private AI labs for national security tools creates inverted supply-chain vulnerabilities. The "risk" designation appears less about foreign influence or technical flaws and more a lever to compel alignment on guardrails — a pattern suggesting intelligence agencies like the NSA are embedding deeply into Anthropic's infrastructure for vulnerability scanning and beyond, while public posturing maintains deniability. This integration risks entangling private AI development with classified priorities, potentially shaping future model training, safety policies, and release strategies in opaque ways that few outlets are scrutinizing. As frontier models like Mythos become strategic assets, the blurred lines between commercial labs and the surveillance state highlight how supply-chain rhetoric masks profound influence over heterodox AI ethics and the accelerating militarization of artificial intelligence.