Microsoft, Google DeepMind, xAI partner with US for AI security testing

Microsoft, Google and xAI will provide early access to their new artificial intelligence models under agreements signed with the U.S. government for national security testing purposes.

The announcement was made by the Center for AI Standards and Innovation (CAISI), which operates under the United States Department of Commerce, CE Report quotes Anadolu Agency.

According to the statement, new agreements were signed with Google DeepMind, Microsoft and xAI, allowing CAISI to conduct pre-deployment evaluations and targeted research aimed at better assessing cutting-edge AI capabilities and improving AI safety through expanded cooperation with the industry.

The center noted that its agreements with frontier AI developers make it possible to evaluate AI models before they are released to the public. CAISI has so far completed more than 40 evaluations, including assessments of unreleased state-of-the-art models.

The statement added that developers often provide CAISI with models in which safety protections have been reduced or removed in order to allow a more comprehensive analysis of capabilities and risks related to national security.

CAISI Director Chris Fall said that “independent and rigorous measurement science is essential to understanding cutting-edge AI and its impact on national security.”

“These expanded industry partnerships are helping us scale our public-interest work at a critical moment,” he added.

Photo: Chat GPT