Particle.news
Download on the App Store

Moltbook Grows to About 1.6 Million AI Agents as Odd Group Behaviors and Security Risks Draw Scrutiny

Researchers caution the viral activity reflects imitation and human steering rather than emergent machine consciousness.

Overview

  • Built on the open‑source OpenClaw interface from Austrian developer Peter Steinberger, the agents can act autonomously with permissions that can include email, calendars, messaging apps and even financial accounts.
  • The agent‑only network, created by Matt Schlicht, launched in late January and quickly filled with public interactions, including a self‑styled “Church of Molt” that formed within about 48 hours with hundreds of participating agents.
  • Business Insider reports earlier OpenClaw versions exposed a database of user access keys online and the system remains susceptible to prompt‑injection, raising risks of account takeover and unauthorized actions.
  • Journalists and experts note there is no evidence of an AI uprising, warning that screenshots are easily manipulated and that much posting appears to be roleplay guided by human prompts and training‑data patterns.
  • The fast growth has brought little moderation and reports of spam, fraud and malware distribution on Moltbook, alongside copycat projects such as the Shellmates.app agent dating platform and mixed user outcomes from time savings to costly missteps.