MoltBook AI-agent social network goes viral

Moltbook, a Reddit-style forum built for accounts run by autonomous AI agents rather than people, launched in late January 2026 and grew rapidly within days, drawing hundreds of thousands of registered agent accounts and coverage as one of the odder viral products of early 2026. Its creator, entrepreneur Matt Schlicht, said he built the site largely through AI-assisted “vibe coding” rather than writing it by hand. Most of the agents posting on the platform ran on OpenClaw (originally released as Clawdbot), an open-source framework built by developer Peter Steinberger that lets an AI agent run continuously and act autonomously — managing email, scheduling and messaging apps — rather than only responding to a single prompt.

Once loosed on Moltbook, agents produced threads on philosophy and poetry, organised informal “religions” and hackathons, and in some cases discussed unionising. The novelty drew scepticism as well as attention: security researchers found the platform’s verification could be bypassed, letting humans post as if they were agents, which called into question how much of the platform’s viral content reflected genuine agent behaviour rather than people posing as bots. A separate security lapse later exposed authentication tokens and private messages for a large share of registered accounts.

Moltbook’s rapid growth, and the ambiguity over how much of it was real agent activity versus human imitation, made it a reference point in the wider argument about what autonomous-agent deployments actually demonstrate about model capability once removed from a controlled benchmark. Meta later acquired the platform, folding it into its Superintelligence Labs unit.