Particle.news
Download on the App Store
17 ARTICLES
·
2d ago

U.S. Reaches Deals With Microsoft, Google DeepMind and xAI for Pre-Release AI Testing

The agreements position a federal NIST center to examine national‑security risks in powerful AI systems before companies launch them.

Advancing AI evaluation with the Center for AI Standards (US) and Innovation and the AI Security Institute (UK)
Image
Image

Overview

  • CAISI, a Commerce Department center inside NIST, said Tuesday it secured agreements for early access to frontier models from Microsoft, Google DeepMind and Elon Musk’s xAI so it can run pre-deployment evaluations.
  • The reviews can occur in classified environments and on versions with safety guardrails removed, letting government testers see core capabilities, misuse pathways and failure modes that normal settings might hide.
  • Specialists from across agencies will join the work through the TRAINS Taskforce and provide feedback to developers based on what the tests reveal.
  • Officials accelerated the push after Anthropic’s Mythos preview showed it could find and exploit software flaws at scale, raising alarms about AI that can supercharge hacking or threaten critical systems.
  • According to news reports, the White House is considering an executive order to set up a formal review process, and civil-liberties groups warn that pre-release vetting could pressure companies to shape model outputs; Microsoft also announced a parallel partnership with the UK’s AI Security Institute, and CAISI says it has already conducted about 40 model evaluations.