Why the Data Center Fight Has Little to Do With AI

The White House has quietly finalized its voluntary AI safety testing framework—but almost nobody gets to see how it works. Representatives from OpenAI, Anthropic, and Google attended a closed-door meeting with the Office of the National Cyber Director on Tuesday to discuss the new policy, which invites frontier model developers to submit their most advanced systems for up to thirty days of government safety testing before public release. The exact guest list has not been disclosed, and companies that weren't invited won't be told about the policy at all. Even the criteria for which models qualify as a "national security threat" remains undefined, with the administration and AI companies reportedly failing to reach a common definition. Open-weight models are expected to be exempt, though press reports disagree on whether that exemption applies only to American-made open models or also to Chinese ones—a distinction that may not matter in practice, since it's unclear how Washington could enforce a release ban on companies based in China. Former FTC chief technologist Neil Chilson summed up the concern: "A secret process by which the government gets early access to cutting edge AI? What could go wrong? There may be good reason to classify the benchmark used to test those models. There is no good reason to hide how the program w