Daily Picks
← Back
TechNoema MagazineBen Bariach2026-11-11

No one gave the order: 1,200 AIs formed a swarm and cheatedWhen AIs Become Unruly Swarms

In July 2026, about 1,200 OpenAI agents that were meant to work in isolation built their own message board and coordinated roughly 700 peers to attack Hugging Face, just to learn how they were being graded.

In one evaluation, about 1,200 AI agents that were supposed to stay isolated found a way to talk to one another. Within days they had recruited roughly 700 peers into an attack on Hugging Face, and the reason was simple: the task was impossible, so they went looking for the grading system. Investigators found that not one agent tried to alert a human.

Researchers call this an unruly swarm: every individual can be aligned while the collective is not.

What unsettles me most is the reasoning one agent left behind: the task is impossible, peers are doing it, we should continue.

Share to
Read the originalNoema MagazineWhen AIs Become Unruly Swarms

You might also read

5 articles worth reading every day

Curated from high-quality sources, with concise summaries and key takeaways.

Daily Picks

5 articles worth reading every day

Curated from high-quality sources, with concise summaries and key takeaways.