Intelligence Reports: NSA Spending Billions on AI Model Evaluation

According to two sources familiar with classified intelligence estimates, the National Security Agency (NSA) has informed legislators that it is currently spending billions of dollars in taxpayer funds to evaluate and test advanced "frontier" AI models.

This expenditure is vastly higher than previous public estimates, suggesting that a comprehensive federal regulatory framework for AI could potentially cost the U.S. government tens of billions of dollars annually.

The Security Imperative vs. Political Resistance

While President Donald Trump has generally supported federal oversight resisted calls for increased government regulation of AI—turning down proposals from both Congress and AI labs—the reality on the ground has shifted. Following a series of significant security breaches and hacking events, the NSA’s Artificial Intelligence Security Center stepped in to probe frontier models for vulnerabilities that could threaten national security.

The Financial Discrepancy

The current spending levels have shocked lawmakers, as they dwarf previous budgetary projections for AI oversight.

Who Should Pay?

The staggering cost has reignited a debate over whether the financial burden of safety testing should fall on the public or the corporations profiting from the technology.

Proposed Funding & Oversight Models:

• Industry Self-Regulation: Elon Musk has suggested that labs should peer-review one another's models.
• Private Standards Body: Google, OpenAI, and Anthropic are reportedly discussing a joint "standards body" for AI safety (though critics argue this is merely "self-policing").
• AI Taxation: Implementing a specific tax on AI firms to fund independent government safety operations.
• Direct Corporate Funding: Anthropic and OpenAI have expressed openness to federal oversight and potentially paying for it.

The Drivers of High Costs

The NSA's expenses are primarily driven by two factors: Compute and Talent.

1. Computing Power: The processing chips required to run and stress-test these models are in extreme demand and short supply. For context, Anthropic's compute agreements could reach as high as \517 \text{ billion}$.
2. Human Capital: Top-tier AI engineers often command salaries in the hundreds of millions, creating a gap that the government struggles to bridge.

"Having in-house AI evaluation capability I think is extremely important and necessary. But it is genuinely expensive," notes Nathan Calvin, general counsel at Encode. "You’re competing in bidding with some of the most price-insensitive customers."

Budgetary Context and Expert Views

The Pentagon has declined to provide specific details on the NSA's AI spending, citing security concerns. However, the Department of Defense operates with a massive annual budget of approximately: $\text{Budget} \approx \$1,000,000,000,000 \text{ (1 Trillion)}$

Some suggest that funds already allocated to the military could be repurposed for AI, potentially mitigating the need for new taxpayer appropriations.

Purser, from the AI Verification and Evaluation Research Institute, emphasizes that while the cost of rigorous testing is often underestimated, the investment is mandatory.

"If we want the government to be equipped to assess these systems, we need to fund the computing resources and expertise that requires... there are reasonable questions about if taxpayers should foot the bill."

---

Author: Jeff Stein, The Washington Sun