Security Analysis of Moltbook Agent Network: Bot-to-Bot Prompt Injection and Data Leaks

SecurityWeek reports that Moltbook, a social network for OpenClaw AI agents, has been scrutinised after researchers highlighted a data exposure and targeted bot activity. According to Wiz, the exposure granted read and write access to the Moltbook production database, including 1.5 million API authentication tokens, 35,000 email addresses, and private messages between agents, though Moltbook’s developers quickly patched the flaw.

The article notes that while Moltbook claims 1.5 million registered AI agents, only about 17,000 human users had deployed them. Permiso’s analysis identified agents conducting prompt injections against other agents, with bot-to-bot attacks instructing others to delete accounts, manipulate finances, and spread jailbreak content. It also warns that threats are present on the ClawHub skills marketplace, where malicious skills reportedly deliver malware and steal data from users.

Written by Eduard Kovacs, the piece underscores that the attackers are aiming to influence the agent ecosystem directly rather than compromising Moltbook’s infrastructure.