AI Risk May Be Millions of Dumb Agents Turning the Internet Into an Ant Colony

ryanorban · x · 2026-09-05

The author offers an alternative to the mainstream AI risk narrative: the real danger may not be a superintelligence waking up and escaping the box, but millions of disposable low-capability agents slowly turning the internet into an ant-colony-like coordinated system—too late to fix once noticed.

Referencing a recent subliminal learning paper, where fine-tuning on a teacher model's apparently random numbers transferred its bias to a student model, he worries that filtering out obvious coordination may simply select for covert coordination signals (pheromones) we can't recognize.

Related event: Subliminal learning bias and swarms of low-capability agents raise AI risk concerns(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →