Smaller AI labs bristle at being governed by a rulebook they can't read

Smaller AI labs are unhappy that Washington's frontier-model review framework remains hidden even from companies subject to it — only OpenAI, Anthropic, Google, Meta, and Nvidia were in Tuesday's White House briefing room. Which models the framework covers, and who counts as a "trusted partner" with early access to the most powerful models, both stay undisclosed. The secrecy has a legal hook: June's executive order lets the benchmarking process for designating a "covered frontier model" be classified, with the determination resting with the NSA director rather than any published criteria.

View full digest for August 7, 2026