When millions of AI agents meet

Short summary

DeepMind researcher Nenad Tomašev explores the future of millions of AI agents operating in coordinated economies—delegating, negotiating, and transacting autonomously. The conversation covers theoretical frameworks for agentic systems, automation psychology, and cybersecurity challenges like agent traps and dynamic cloaking. Key focus: building trustworthy distributed intelligence while managing cognitive monoculture risks.

• •AI agents will form autonomous economies where they delegate and negotiate with each other
• •Critical security and psychological risks emerge: automation bias, agent traps, cloaking attacks
• •Building trustworthy distributed intelligence requires new frameworks for safety and coordination

Generated with AI, which can make mistakes.