US to test next-generation AI with Google DeepMind Microsoft xAI
The US is working with Google DeepMind, Microsoft, and xAI to put next-generation AI models through some serious testing.
Announced by the National Institute of Standards and Technology (NIST), this move is all about pre-deployment evaluations, targeted research, and post-deployment assessments to improve understanding of these powerful AIs and their potential national security implications.
CAISI pushes models past safety limits
The Center for AI Standards and Innovation (CAISI) will run pre-release tests on these advanced models, even pushing them without usual safety limits to spot any weak points early.
They have already done over 40 assessments and will keep checking after launch.
This effort brings together multiple agencies through the TRAINS task force, launched in 2024 to focus on keeping America's AI future secure.
Chris Fall stresses measurement science
CAISI Director Chris Fall summed it up: "rigorous measurement science remains necessary for understanding frontier AI systems and their broader implications."
These partnerships build on earlier partnerships announced in 2024 and fit into the bigger picture of America's push for responsible AI development.