Classified Estimates Show the NSA Is Paying Billions to Test AI Models
3 hours ago
- The NSA is spending billions of dollars in taxpayer funds this year to evaluate and test advanced AI models, according to classified intelligence sources.
- The high cost has led lawmakers to believe a comprehensive AI regulatory system could cost tens of billions of dollars annually.
- President Trump has resisted federal AI oversight, but recent hacking incidents prompted NSA's AI Security Center to test frontier models for national security vulnerabilities.
- The expensive efforts are intensifying debates about government's role in AI review and may push frontier labs to bear the financial burden.
- Previous regulatory proposals were far cheaper: the AI Security and Innovation Act would cost ~$20 million/year, and another bill would cost $36 million over five years.
- Some tech leaders like Elon Musk suggest labs should review each other's models; Google, OpenAI, and Anthropic are jointly creating a safety standards body.
- AI safety experts reject self-policing, proposing a government tax on AI companies to fund independent safety audits and reduce taxpayer burden.
- NSA's AI testing is funded by classified national security budget; exact spending is unclear, and the Pentagon declined to comment.
- Major AI costs include extra computing power for testing (e.g., chips) and hiring personnel, with top engineers' salaries reaching hundreds of millions.
- The Pentagon's budget is nearly $1 trillion, and some military funds could be shifted to AI spending.
- Experts urge swift third-party audits and government investments, with possible assessments on frontier developers to fund independent audits.