Google DeepMind funds $10M multi-agent safety research initiative

DeepMind warns that millions of autonomously-interacting agents enable scams, cyberattacks and prompt injection at internet scale, and is funding a $10M program to study the risks.