Classified Estimates Show the NSA Is Paying Billions to Test AI Models (www.washingtonsun.com)

The National Security Agency (NSA) has revealed to lawmakers that it is investing billions of dollars this year in assessing advanced AI models, a figure much higher than previously estimated. This significant spending prompts concerns that establishing a comprehensive AI regulatory framework could cost the government tens of billions annually. The NSA's increased focus on evaluating AI stems from the need to address potential national security vulnerabilities highlighted by recent hacking incidents, yet this raises questions about who should bear the financial burden—taxpayers or the tech companies developing the models. The escalating costs primarily arise from the need for advanced computing power due to high global demand and talent acquisition, as the NSA competes with private sector firms for top AI engineers. Current budget estimates suggest that the costs might be funded through classified military budgets, complicating oversight and accountability. Experts argue for the urgency of implementing independent third-party audits and even suggest that AI developers could be taxed to support regulatory efforts, allowing the government to maintain oversight without placing the financial burden solely on taxpayers. The debate intensifies as corporations like Google, OpenAI, and Anthropic explore self-regulatory standards, though many safety experts warn that self-policing may not suffice for effective oversight.