White House Builds Classified Cyber Tests for Frontier AI While Keeping Reviews Voluntary
The U.S. government is building a classified benchmark for advanced AI cyber capabilities while keeping its pre-release review framework voluntary; White House advisers have also told developers that open-weight models will not be included in the voluntary testing regime.
Why it matters: The framework moves frontier-model cyber capability closer to the center of enterprise AI risk management while leaving a major policy gap around open-weight systems.
The U.S. government is formalizing a new security layer around increasingly capable artificial-intelligence models. A June executive order directed federal agencies to create a classified benchmarking process for advanced cyber capabilities and a voluntary framework through which developers can provide secure pre-release access to covered frontier models for up to 30 days.
The order explicitly stops short of mandatory licensing, preclearance or permitting for the development or release of new AI models. Reuters reported this week that White House advisers told AI companies that open-weight models will not be put through the voluntary safety tests, creating a significant distinction between closed frontier systems and models whose weights can be downloaded or redistributed.
For enterprises, the policy signals that model capability itself is becoming part of cybersecurity governance. As AI systems become more capable of vulnerability discovery, exploitation and autonomous action, pre-deployment testing, model access controls and auditability are likely to become more important procurement criteria.