The Center for AI Standards and Innovation (CAISI) has been ordered to halt all public evaluations of frontier artificial intelligence models. Signed on June 2, 2026, the new executive order shifts control from open scientific assessment to a classified national security framework managed by the Treasury Department, the National Security Agency (NSA), and CISA.
The directive follows the emergence of Anthropic’s unreleased Claude Mythos model in April 2026. The model autonomously identified thousands of previously unknown high-severity zero-day vulnerabilities across major operating systems, prompting the administration to deem existing oversight mechanisms insufficient.
Previously, CAISI had established itself as a neutral benchmarking authority, completing over 40 public evaluations by early June 2026. It maintained partnerships with Google DeepMind, Microsoft, and xAI based on collaborative transparency. That dynamic ends now.
Under the new regime, all US-based AI companies face a mandatory 30-day government review before releasing any model. While firms will still submit their technology for scrutiny, results will remain classified. This creates a significant competitive disadvantage for American firms compared to international rivals in China and Europe, who operate without such constraints.
The shift marks a transition from open science to national security asset management. Researchers at CAISI now work within a classified apparatus, potentially driving talent toward academia or private sectors where publication remains possible. Independent verification of AI safety claims effectively disappears, leaving market confidence reliant solely on corporate assertions.