Google, Microsoft and xAI have agreed to allow U.S. government review of new AI models for potential national security and cybersecurity risks before the models are widely released. The arrangements involve government vetting of unreleased versions of the systems, following concerns from U.S. officials about how advanced AI could affect cybersecurity. The move comes after heightened attention to AI-related security risks prompted by Anthropic’s latest model, widely reported as “Mythos.” Multiple outlets describe that the reviews are intended to help identify and mitigate vulnerabilities or misuse risks before deployment. In addition to the companies’ participation, the U.S. National Institute of Standards and Technology (NIST) is reported to be involved in the testing framework. The announcements reflect a broader shift toward government oversight of frontier AI capabilities, with the companies working within a structured national security review process. Terms and scope of the reviews were not detailed in the provided summaries, but all reporting indicates a coordinated approach to pre-release assessment of new AI systems by the participating firms.