The US Blacklisted Anthropic as a Security Threat. Its Spy Agencies Are Using Claude Anyway.

Why It Matters

The reliance on Anthropic despite its blacklist highlights a strategic vulnerability: without domestic hardware capacity, U.S. intelligence agencies risk dependence on foreign‑linked AI providers, potentially compromising national security.

Key Takeaways

• Pentagon blacklists Anthropic, but NSA keeps using Claude
• $9 B emergency fund targets Nvidia Grace Blackwell superchips
• Chip shortage forces reliance on private AI models for classified work
• Anthropic’s revenue jumps to $30 B, eyeing $800 B valuation
• New data centers aim to eliminate single‑vendor AI dependency

Pulse Analysis

The clash between policy and practice has placed the NSA in an uncomfortable position. While the Department of Defense labels Anthropic a supply‑chain threat due to its foreign‑investment ties, the agency’s need for advanced reasoning capabilities leaves it with Claude as the only model that can meet the speed and accuracy required for real‑time intelligence analysis. This pragmatic exception, authorized by White House chief of staff Susie Wiles, underscores how operational imperatives can override formal security directives when alternatives are unavailable.

At the heart of the dilemma is a broader semiconductor crunch that has rippled through both consumer and government sectors. Advanced AI models demand the latest high‑bandwidth chips, and the current scarcity of Nvidia’s Grace Blackwell superchips forces the intelligence community to seek emergency funding. The $9 billion package, part of an $800 million fast‑track allocation, will finance custom‑built, liquid‑cooled data centres capable of handling the massive power draw of frontier AI workloads. By securing a domestic hardware pipeline, the government hopes to regain control over its AI stack and reduce reliance on any single vendor.

Long‑term, the episode reveals a structural tension in U.S. AI strategy. As frontier models concentrate in a handful of private firms, the ability of regulators to dictate supply‑chain choices diminishes, especially when the hardware to run competing models is lacking. Investing in sovereign semiconductor capacity and flexible data‑center infrastructure could decouple national‑security operations from private AI providers, preserving both technological edge and supply‑chain integrity. Until such capabilities are in place, agencies like the NSA will likely continue to balance security mandates against the practical need for the most capable AI tools available.