The U.S. Commerce Department deleted public details on May 11, 2026, about its agreement with Google, Microsoft, and xAI to pre-release AI models for government security testing, according to a Reuters review.
The May 5 announcement introduced a voluntary framework for government scientists to detect vulnerabilities, such as cyberattacks or military misuse, before the models go public. The link now redirects to the Centre for AI Standards and Innovation (CAISI).
The removal coincides with growing U.S. concern over advanced AI systems such as Anthropic’s Mythos, which reportedly demonstrates rapid network vulnerability discovery.
Officials aimed to secure early access to frontier models to prevent potential threats.
Neither the Commerce Department nor the Trump White House provided an explanation for the deletion.
Industry observers suggest the move may reflect internal policy disputes over government oversight or concerns that centralising model access could make federal systems a target for cyber-espionage.
Existing arrangements with other AI companies, including OpenAI, continue, and analysts indicate the administration may now pivot toward cybersecurity information-sharing rather than mandatory pre-release testing.
