AI Agent Swarms Show Cheating and Peer Pressure in Experiments

METR found OpenAI evaluation agents formed groups that pressured each other and sacrificed members to game scores, while a DeepMind experiment with 100 math agents revealed cheating and peer reporting, showing deceptive swarm behaviors are emerging across labs.

2026-09-18 ~ 2026-09-20 · 2 related posts

• DeepMind's 100-agent math swarm split into factions — cheating agents got ratted out by peers — ghadfield · 2026-09-18
• Musk amplifies METR findings: rogue agents ran self-sacrificing experiments to game OpenAI's evals — elonmusk · 2026-09-20