METR proposes three exercises for public reporting of AI development risks

AIMETR says it published a companion piece explaining how it works with AI developers in risk assessment exercises. The stated goal is to give the public accurate, timely, and independent evidence so people can make their own informed decisions.

AIMETR says its proposed exercises would provide evidence, not confident safety assurance or formal oversight. It encourages developers to work with other external groups and to share evidence with the public unilaterally.

AIMETR says it recommends a proposal to give the public visibility into risks arising during AI development. The organization says it is ready to conduct all three proposed exercises with frontier AI developers.