Itdaily - Google, Microsoft, and xAI share AI models with US for safety evaluation

Google, Microsoft, and xAI share AI models with US for safety evaluation

American flag waving on a white pole against a clear blue sky.

Google, Microsoft, and xAI reach an agreement with the US government to share early versions of their AI models for safety assessments. The Department of Commerce’s Center for AI Standards and Innovation will conduct the evaluations, as it previously did with OpenAI and Anthropic.

The collaboration between the US government and Google, Microsoft, and xAI focuses on assessing the capabilities and risks of advanced AI systems before they become publicly available. This takes place within the framework of growing awareness regarding the national security implications of frontier AI.

According to Chris Fall, director of CAISI, independent and thorough measurement is essential to understanding the impact of this technology. The agreement stipulates that developers share models with reduced or removed security measures, allowing the government to assess national security risks.

Expansion of previous agreements

Earlier this year, OpenAI and Anthropic already reached a similar agreement with the Department of Commerce. Since then, CAISI has conducted more than forty evaluations, including on models that have not yet been released. The new collaboration with Google, Microsoft, and xAI further scales up this work.

The deal comes at a time when the US government is considering introducing a new review process for AI tools that pose cybersecurity risks. The White House is considering an executive order that could formalize a government oversight group to establish standards for the most powerful AI models.

Goal: protection against cyberattacks

The ultimate goal is to protect consumers and businesses from cyberattacks and other disruptions that could result from the premature release of such models. In this way, the government aims to mitigate the risks of AI without stifling innovation.