Core Summary

  • The AI Standards and Innovation Center (CAISI) under the U.S. National Institute of Standards and Technology (NIST) has signed agreements with Google DeepMind, Microsoft, and xAI to conduct pre-deployment evaluations and research on their frontier AI capabilities.
  • The agreements build on previously announced partnerships with OpenAI and Anthropic, aiming to strengthen AI safety, and have been incorporated into directives from CAISI, the Commerce Secretary, and the AI action plan issued by President Donald Trump last year.
  • The agreements allow the government to conduct evaluations before cooperative AI models are made available to the public, as well as ongoing assessments after deployment. CAISI Director Chris Fall said, "Independent, rigorous measurement science is essential for understanding frontier AI and its national security implications."

In-Depth Analysis

As the primary government liaison for the tech industry, CAISI and its industry partnerships support information sharing, product improvements, and help understand current and future AI capabilities in the U.S. and abroad, NIST stated. These collaborations also enable NIST to gain insight into the national security capabilities and risks that AI models may pose.

Trump initially adopted a hands-off approach to AI regulation in the first year of his second term, aiming to accelerate AI innovation, build domestic AI infrastructure, and advance international AI diplomacy and security. However, The New York Times reported on Monday that the administration is seeking to strengthen AI oversight, as policymakers face pressure from national security officials regarding risks posed by powerful AI models such as Anthropic's Mythos. Initiatives like Project Glasswing (Anthropic's effort to identify and fix software vulnerabilities) further highlight a growing trend of expanding governance alongside technology adoption.

The expansion of the CAISI framework indicates that "sovereign alignment" will become a mandatory criterion in enterprise AI procurement, Nick Patience, vice president and practice lead for AI at The Futurum Group, said in an email to CIO Dive.

In March, the Department of Defense formally designated Anthropic as a security risk, a decision upheld by a federal judge last month, despite the company's participation in CAISI's evaluation process. Patience noted that this demonstrates suppliers can still face sanctions if their internal ethics conflict with national security directives.

For chief information officers (CIOs), the new agreements with Google, Microsoft, and xAI serve as a form of "political insurance," Patience said. Choosing a vendor that has not received favored status from the Commerce Department and NIST, especially for enterprises with or seeking federal contracts, poses a "significant contagion risk."

"We have entered an era where a model's utility to the nation is a key predictor of its long-term viability in enterprise technology stacks," he said.