White House keeps advanced AI testing framework confidential

By choosing not to publish its advanced AI testing framework, the White House is signaling that it views parts of model evaluation as national security‑sensitive. The framework will govern how companies work with the US government to assess whether under‑development models fall under special controls, and how early the state can get access to pre‑release systems. Keeping the details classified might make sense for some cyber capabilities, but it also leaves researchers, allies and smaller firms guessing about what the US considers unacceptable risk.

Strategically, this deepens the split between a small circle of frontier labs with privileged access and everyone else. If Nvidia, a few hyperscalers and select model developers get to co‑design the benchmarks that matter for US policy while open‑weight and smaller players are excluded, the framework itself becomes a lever of industrial policy. It could also complicate coordination with Europe and other allies who are moving toward more transparent AI safety regimes under instruments like the EU AI Act.

For the AGI timeline, the framework is a double‑edged sword. If it is well designed and consistently applied, it could reduce the odds that a dangerously capable model slips into production without adequate red‑teaming. If it is opaque and narrowly shared, it risks becoming a black box that slows outside scrutiny while doing little to constrain bad actors abroad.