The program for evaluating «sufficiently advanced» models by the administration will now be reviewed by AI labs in a staff meeting in DC this Tuesday.
Anthropic, OpenAI, Google and Meta are expected to attend, Politico writes, with the White House saying they are talking to «many more» industry partners, Axios reports.
The idea of the framework in the original order was intended to give the government 30 days prior to release of «sufficiently advanced» models to do extensive testing for possible dangers, as seen lately with cyber capabilities.
The Fable and Mythos models and GPT-5.6 were all delayed in June after government intervention, and this framework is supposed to offer a regular, voluntary way to whitelist models as safe enough for release, or flag dangers to be improved.