The White House finalized its frontier-model testing framework on August 4 and exempted open-weight models entirely.

About a dozen companies including OpenAI, Anthropic, Google and Meta met White House staff in a roughly 30-minute session closing two months of negotiation from Trump's June AI cybersecurity executive order. Administered by CAISI inside NIST, it asks developers of covered frontier models (defined as closed-source, state-of-the-art, national-security-relevant) for a 30-day early access window before public release so the government can run cybersecurity evaluations. Open-source and open-weight releases face no review. That's a codified competitive asymmetry: five closed labs absorb release friction, everyone else ships free. Given the week's evidence that the risk lives in eval harness configuration rather than weights, I'm not sure the framework is even pointed at the right thing.