White House Finalizes AI Cybersecurity Testing Framework While Keeping Its Contents Private
The White House has finalized a voluntary framework requiring major AI developers to grant the government up to 30 days of pre-release model access for cybersecurity evaluation, but the framework itself has not been publicly released and the administration has not disclosed testing metrics, reporting requirements, or which agency leads the effort. The classified benchmark that determines which models qualify as 'covered frontier models' remains undisclosed, as do the government's review conclusions and any public-results process. Officials have suggested framework participation may become a condition for federal funding eligibility, including Defense Department contracts, though no published rule confirms this.