Official summary

Our first public, evidence‑based assessment of how the world’s most advanced AI systems are evolving, bringing together results from two years of AISI's frontier model testing.

The record

Published by

AI Security Institute