Microsoft, Google and xAI agree to U.S. government pre-launch testing of AI models

Microsoft, Google and xAI have reached agreements to allow U.S. government agencies to test unreleased AI models for cybersecurity and national-security risks before those models are publicly launched, the Center for Security and Emerging Technology at Georgetown University reports. CSET analyst Jessica Ji shared her perspective in a commentary carried by CNN and reposted by CSET.

The arrangements are voluntary and aimed at giving federal evaluators early access to model behavior and potential vulnerabilities that could affect national infrastructure and security. CSET’s write-up frames the company-government testing as part of a broader push to manage AI risk while companies continue to deploy powerful systems.

Emerging Technology

CSET, based at Georgetown University in Washington, D.C., highlighted the significance of these cooperative steps as the federal government sharpens its approach to AI oversight. The piece notes the testing is intended to address cybersecurity concerns, evaluate misuse risks, and inform potential mitigation strategies prior to public release.

Experts at CSET emphasize that allowing pre-launch access creates an opportunity for government specialists to identify risks early, but they also flag the need for clear protocols on information sharing, scope of access, and legal safeguards. The commentary situates these company agreements within ongoing policy debates in Washington over how to balance innovation, commercial confidentiality and national security.