New research proposes a framework for evaluating general-purpose models against novel threats

Source: https://deepmind.google/discover/blog/an-early-warning-system-for-novel-ai-risks/