The U.S. government is intensifying its oversight of artificial intelligence by establishing new agreements with leading developers, including Google DeepMind, Microsoft, and xAI. The National Institute of Standards and Technology (NIST) announced that its Center for AI Standards and Innovation (CAISI) will now conduct systematic evaluations of advanced AI models prior to their public release, focusing specifically on national security risks.
CAISI is set to become the pivotal link between the government and the AI industry. Under the direction of Commerce Secretary Howard Lutnick, this initiative will coordinate testing, joint research, and the development of standards for commercial AI systems.
A significant aspect of these new agreements is the government's access to AI models before they are released to the public. This early access allows for independent assessments of capabilities and potential threats, with over 40 evaluations already conducted, including assessments of cutting-edge systems that have not yet been made public.
Chris Fall, the director of CAISI, emphasized the importance of rigorous and independent metrics for understanding the capabilities of advanced AI and its implications for national security. He noted that these new agreements come at a critical time as competition in the AI sector is rapidly escalating.
The testing process will mimic real-world risk scenarios, allowing developers to submit versions of their models with reduced protective measures to gauge potentially dangerous behaviors. Experts from various agencies will participate in these evaluations, collaborating in an interagency group known as TRAINS, which focuses on AI and national security.
Additionally, the agreements facilitate data sharing between the government and companies, aimed at expediting product refinement and enhancing transparency in evaluating their capabilities. Some of the research will take place in controlled environments, including classified settings.
This comprehensive approach to AI oversight signals a significant shift in how AI technologies will be developed and assessed, potentially reshaping the competitive landscape for AI developers and ensuring greater accountability in the industry.
Informational material. 18+.
