Microsoft to Provide Early AI Model Access for 40+ U.S. Government Evaluations

MSFTMSFT

Microsoft, Google and xAI will grant the U.S. Center for AI Standards and Innovation early access to their advanced AI models for national security risk assessments, marking over 40 completed evaluations including unreleased models. This follows 2024 agreements with OpenAI and Anthropic as scrutiny intensifies over AI risks.

1. Partnership for Early AI Access

Microsoft has joined Google and xAI in a formal agreement to share its most advanced AI models with the U.S. Center for AI Standards and Innovation (CAISI). The collaboration grants government evaluators early-stage access to these models, aiming to identify potential national security risks before public release.

2. Scope and Scale of Risk Evaluations

CAISI has completed more than 40 assessments of advanced AI systems, including proprietary and unreleased versions from leading tech firms. These evaluations examine factors such as model robustness, potential misuse scenarios and alignment with ethical guidelines to inform risk mitigation strategies.

3. Industry Precedent and Regulatory Outlook

This latest deal builds on similar 2024 arrangements with OpenAI and Anthropic, reflecting growing demand for standardized AI oversight. By engaging government bodies early, Microsoft and peers seek to shape emerging regulations and demonstrate commitment to safe AI deployment.

Sources

FFBBB
+1 more