DeepMind's $10M bet on preventing multi-agent anarchy

google deepmind has committed $10 million to academic research on multi-agent safety, a field rohin shah says is virtually non-existent despite agents being a centerpiece of google i/o. the initiative, backed by schmidt sciences and the uk government, aims to simulate millions of agent interactions to forecast emergent risks like supercharged cyberattacks and market manipulation before agents permeate the economy. shah warns that studying agents in isolation is insufficient and that the industry has only a few months to understand these dynamics before deployment hits critical mass.