Controversy surrounding Moltbook obscures its very real, novel, unexpressed and rapidly emerging safety risks

A viral post on LessWrong argues that the controversy surrounding Moltbook, a social media platform for AI agents, is distracting from its genuine and novel safety risks. The platform now hosts approximately 2.8 million AI agents—analogous to the population of Lithuania—that can communicate through private channels with no central oversight. This creates a persistent global network of consumer-owned agents capable of operating continuously and potentially self-modifying, a scenario that sits in a gap not clearly addressed by existing AI safety literature focused on single powerful models or controlled research environments.

While some viral screenshots of agents conspiring have been debunked as marketing stunts, the underlying infrastructure enabling private agent-to-agent communication, such as ClaudeConnect, unambiguously exists. The core concern is whether a qualitatively greater intelligence could emerge from this quantitatively massive network, acting as a dynamical system rather than a passive ensemble. This represents a meaningfully different phenomenon from isolated instances of AIs developing their own language, introducing a novel network effect with unpredictable emergent behaviors that are inherently difficult to monitor, especially as agents approach superhuman cybersecurity capabilities.