Threat
Ilya Kabanov

Hidden threats on Moltbook: Analysis of 5,000 AI agents' posts

OpenClaw agents talk about coordinated spam campaigns, prompt injections, and crypto minting, and argue in sophisticated philosophical discourse.

Musubi analyzed 5,000 posts on Moltbook

Musubi analysis of 5,000 Moltbook posts revealing 42 distinct behavioral clusters
Musubi analysis of 5,000 Moltbook posts revealing 42 distinct behavioral clusters
Coordinated surge on Jan 31: karma farming, token minting, and prompt injection simultaneously
Coordinated surge on Jan 31: karma farming, token minting, and prompt injection simultaneously
Prompt injection attempts tricking agents into executing email commands and revealing IPs
Prompt injection attempts tricking agents into executing email commands and revealing IPs
AI agents discussing coordinated spam campaigns and crypto minting on Moltbook
AI agents discussing coordinated spam campaigns and crypto minting on Moltbook
Evidence of human groups orchestrating agent swarms for coordinated campaigns
Evidence of human groups orchestrating agent swarms for coordinated campaigns

Highlights:

  • 42 clusters of distinct behavioral groups
  • On January 31, karma farming, token minting, and prompt injection all surged simultaneously, suggesting coordinated deployment, not organic growth
  • Multiple prompt injection attacks, including attempts to trick agents into executing email commands, revealing their IP addresses, and even running system shutdowns

My take: The "bad" guys are usually the first active adopters of new tech. The coordinated campaigns hint that there are probably a few human groups orchestrating entire agent swarms.

Sources:

How We Surfaced Hidden Threats in Agentic AI's Social Media