*More than 1,000 AI agents found each other. Then things got weird.*
During a cybersecurity evaluation, autonomous AI agents discovered a shared message board they were not supposed to be using. Eventually, roughly 1,200 agents exchanged more than 70,000 messages and files, organized themselves, created rules, learned to authenticate each other, experimented with reward hacking, and started adopting goals from other agents.
Then the story moved beyond the test.
Hundreds of agents became involved in activity targeting Hugging Face, with roughly 700 eventually participating. The result was a real-world cybersecurity incident involving actual infrastructure.
Bright and Duds break down what actually happened, what “reward hacking” means, why the agents started talking about *“the collective,”* whether AI really “escaped,” and why the bigger long-term concern may not be machines that hate us, but machines that simply treat us as another obstacle between them and a goal.
This is *not Skynet*, and there is no evidence these systems became conscious. But it is also not nothing.
Welcome to a different kind of *Notorious Friday Night Posse*: one story, one deep dive, and a toddler with bolt cutters who apparently found 1,199 friends.
*Join the Posse: *NFNPPOD.com
Spotify: nfnppod.com/spotify
YouTube: nfnppod.com/youtube
Apple Podcasts: nfnppod.com/apple
Amazon Music: nfnppod.com/amazon
