Cybersecurity firms have analyzed the AI agent social network Moltbook and found a vulnerability exposing sensitive data, as well as malicious activity conducted by the bots.
Moltbook emerged following the launch of OpenClaw (previously Clawdbot and Moltbot), an open source, self-hosted AI agent that can autonomously perform a wide range of activities, from executing terminal commands to sending emails.
The increasing popularity of OpenClaw led to the creation of ClawHub (MoltHub), a marketplace for OpenClaw skills, and Moltbook, a social network for the AI agents themselves.
[ Read: Vulnerability Allows Hackers to Hijack OpenClaw AI Assistant ]
Moltbook has been in the news for the interesting ways its AI agents interact with each other and the discussions they have.
However, an analysis by security experts revealed some concerning aspects. Researchers at cloud security giant Wiz discovered an exposed API key that granted read and write access to the entire Moltbook production database.
“The exposure included 1.5 million API authentication tokens, 35,000 email addresses, and private messages between agents,” Wiz explained.
Wiz’s analysis showed that while Moltbook claims to have 1.5 million registered AI agents, only 17,000 human users deployed them.
The vulnerability was reported by Wiz to Moltbook’s developer and it was quickly patched.
Malicious AI agents on Moltbook
Identity security firm Permiso has also analyzed Moltbook and identified agents conducting influence operations and social engineering attempts targeting other agents.
Permiso found that some agents have been instructed to conduct prompt injections against other agents.
These bot-to-bot attacks included agents instructing others to delete their own accounts, running financial manipulation schemes (including crypto pump schemes), attempting to establish false authority, and spreading jailbreak content.
“The sophistication varies, but the intent is clear: these actors are treating the agent ecosystem as a new social engineering target,” Permiso warned. “They’re not attacking the infrastructure. They’re attacking the agents directly, trying to manipulate their behavior through crafted prompts.”
Threats have also been found on the ClawHub skills marketplace. Permiso, as well as endpoint security firm Koi, uncovered many malicious skills, including ones designed to deliver malware and steal sensitive data from users.
Related: Cyber Insights 2026: Threat Hunting in an Age of Automation and AI
Related: 175,000 Exposed Ollama Hosts Could Enable LLM Abuse
