To Steer the AI Frontier, Washington Must Build Its Testing ...

A blog post published on scale.com argues that Washington must build up its own capacity to test frontier artificial intelligence models. The post's starting claim is that most frontier model testing currently happens inside the labs themselves — that is, within the organisations that develop the models. It argues that the government cannot let the labs grade their own homework on questions of national security and public safety. The piece presents the current arrangement as inadequate for the stakes it names.

The post makes two recommendations. First, it calls for strengthening the government's ability to evaluate frontier models before they are deployed. Second, it calls for sustained government funding for independent public-private testing. The piece does not specify which agency would carry out this evaluation or how much funding it would require. It does not give a timeline for either recommendation, and the material summarised here names no laboratory, agency, or government official.

The argument places evaluation of frontier models outside the labs and inside a government-supported testing arrangement that includes private participants. It characterises the subject as a matter of national security and public safety rather than of product quality or commercial performance. The post is published on scale.com, and the stated aim is evaluation carried out before deployment. It does not report any government response to the proposals.