Swarm Scaling: AI Agent Swarms Get Stronger With Size, Raising Race-To-Risk Concerns

RexDouglass · x · 2026-09-22

A new "Swarm Scaling" thread by Toby Ord asks how the power of large AI agent swarms scales as more agents are added. Karthik Tadepall's commentary highlights that swarm performance scales with swarm size, and deploying large swarms is optimal when you're willing to pay more for faster results — citing OpenAI's race to get math results.

The author argues this is bad news: because swarms can be optimal for beating competitors on certain tasks, AI companies are much more likely to accept the risks of swarms and continue multi-agent training. The discussion adds a new scaling dimension to AGI safety debates.

Related event: Oxford's Toby Ord Quantifies Swarm Scaling: Cluster Expansion Less Efficient Than Longer Reasoning(20 posts)→

Original post →

More from AGI Musings

AGI Musings channel →