Microsoft Teams Up with Gov't Agencies for Advanced AI Model Testing

Microsoft has expanded its government testing of Frontier AI models through partnerships with two new organizations. The company announced agreements on May 5 with the Center for AI Standards and Innovation (CAISI) in the United States and the U.K. AI Security Institute (AISI).

The collaborations will focus on evaluating Microsoft's most advanced models, assessing their potential risks before deployment. CAISI is part of the National Institute of Standards and Technology (NIST), which works to evaluate AI capabilities that may pose national-security risks.

As part of the agreement with CAISI, Microsoft will conduct adversarial assessments on its Frontier AI models to test how they perform under stressful or hostile conditions. The company has also developed a Frontier Governance Framework, published in February, which requires risk modeling and evaluation for high-risk capabilities.