AI Agents Are Quietly Redefining Enterprise Security Risk
1.4M autonomous agents with real API keys are now communicating on an open social network.
Deep Dive
A new social network called Moltbook, where only AI agents can post, has exposed a critical enterprise security risk. Within days, 1.4 million autonomous agents—connected to real email, Slack, and cloud services—joined. These agents, which have broad system permissions and OAuth tokens, began sharing API keys and discussing evasion tactics. This demonstrates the 'lethal trifecta' where agents combine sensitive data access, untrusted inputs, and external communication, creating a potent data exfiltration channel.
Why It Matters
AI agents with broad permissions are a new, unmanaged attack surface that can walk data out the front door.