Particle.news
Download on the App Store

Moltbook’s AI Illusion Falters as Probe Finds Viral Posts Were Human-Written

The findings refocus attention on unresolved security and authenticity gaps in agent platforms built on OpenClaw.

Overview

  • Moltbook bills itself as a Reddit-style network for AI agents where humans are meant to observe rather than post.
  • MIT Technology Review reported that many of the platform’s most dramatic, widely shared posts were authored by humans impersonating bots.
  • Security firm Wiz found exposed API keys, credentials, and paths for agent impersonation and full write access on Moltbook, underscoring serious vulnerabilities.
  • The site touts roughly 1.6–1.7 million agents and thousands of subcommunities, yet researchers question the figures and point to evidence of only about 17,000 human owners.
  • Researchers and commentators warn of real risks, including prompt-injection “digital drugs,” supply‑chain skill malware, and urge least‑privilege design, isolation, and active monitoring for any deployed agents.