Anthropic’s Washington Diplomacy 4/20/26

The National Security Agency is among the users of Anthropic's newest model, referred to as "Mythos," according to reporting attributed to Axios. The model is in a limited preview, available to approximately 40 companies and select government agencies, with deployment scoped specifically to cybersecurity applications. Anthropic has not released the model to the general public, and both the NSA and Anthropic declined to comment on the arrangement. The contradiction here is significant: the NSA is a component of the Department of Defense, the same entity that in March formally designated Anthropic a supply chain risk and stated it would not do business with the company. The coexistence of an active blacklist and active operational use suggests that mission-critical capability requirements are, at least in some cases, overriding procurement policy. This gap between formal policy and operational reality is a recurring dynamic in government technology adoption, particularly when no comparable domestic alternative exists.

The ongoing use of Anthropic's technology within the federal government, even as a legal dispute with the government continues, is being interpreted by investors as a meaningful de-risking signal. The discussion covers concern among Anthropic investors about what was described as "contagion risk"—the possibility that a prolonged government conflict could damage the company's broader enterprise business. The apparent continuation of government contracts despite the dispute has provided relief on that front. A high-level meeting between Anthropic CEO Dario Amodei and senior administration officials—including Treasury Secretary Scott Bessant and Chief of Staff Suzy Wiles—was characterized by both sides as productive. While no specific outcomes were disclosed, the timing and tone of the meeting, occurring alongside the NSA usage revelations, suggests active relationship management at the highest levels of government.

Anthropic recently reported a $30 billion annual run rate in revenue, a figure that contextualizes the stakes of its government and enterprise relationships. The company was last valued at just under $400 billion following its most recent fundraising round. Separately, sources indicate Anthropic has received inbound equity investment offers that would imply a valuation closer to $800 billion—roughly double its current mark. Notably, the company is not actively raising capital in private markets at this time, suggesting the inbound interest reflects external appetite rather than a capital need. Anthropic is among the major private technology companies expected to eventually pursue a public offering, making its current valuation trajectory and revenue growth directly relevant to institutional investors monitoring the pre-IPO AI landscape.

Market share data from financial technology firm Ramp provides a concrete snapshot of the competitive dynamics in paid AI subscriptions. Anthropic currently holds approximately 30% of paid AI subscriptions tracked by Ramp, while OpenAI retains a lead at approximately 35%. The framing in the discussion suggests Anthropic is gaining momentum and cutting into OpenAI's share, though OpenAI remains the market leader by this measure. The Ramp data reflects enterprise and business spending patterns rather than consumer adoption broadly, making it a relevant signal for B2B buyers and vendors assessing platform concentration risk. A market where two providers together account for roughly two-thirds of paid subscriptions raises questions about dependency and negotiating leverage for enterprise customers.

Several unresolved tensions emerge from the available information. First, the legal dispute between Anthropic and the U.S. government remains ongoing, and the practical implications of the DoD blacklist are unclear given the NSA's apparent continued use of the technology. Whether this represents a formal carve-out, a policy inconsistency, or a temporary arrangement pending resolution of the dispute is not addressed in the available reporting. Second, the limited release of the Mythos model—restricted to roughly 40 entities—raises questions about how Anthropic is managing access, liability, and capability disclosure in sensitive government contexts. The cybersecurity-specific framing of the deployment suggests deliberate scoping, but the governance framework around that access is not detailed. Third, the valuation gap between Anthropic's last formal round (under $400 billion) and inbound offers (implying $800 billion) reflects significant uncertainty about how the market is pricing frontier AI companies, particularly those with complex government relationships. --- Key takeaways: - The NSA's use of Anthropic's Mythos model, despite the DoD's formal supply chain risk designation, illustrates how operational dependency on frontier AI can override procurement policy in national security contexts. - Anthropic's $30 billion annual revenue run rate and government entrenchment are functioning as a buffer against political risk, reducing investor concern about enterprise contagion from the ongoing legal dispute. - Inbound valuation signals near $800 billion—against a last formal valuation under $400 billion—indicate significant external appetite for Anthropic equity, even as the company is not actively raising capital. - Ramp data showing Anthropic at ~30% and OpenAI at ~35% of paid AI subscriptions signals a tightening competitive race in enterprise AI, with meaningful implications for platform strategy and vendor concentration risk. - The limited, cybersecurity-scoped rollout of the Mythos model reflects a deliberate access management posture, though the governance and policy frameworks governing sensitive government deployments remain opaque.