China's CNVD warns malicious AI agent skills for OpenClaw hide trojans that steal credentials

1. May 11, 2026, 12:00 PM
   CNVD publishes advisory^[1]
2. Nov 12, 2026, 7:52 AM
   CVERC warned on 22 April that OpenClaw skill packages had been poisoned with malware^[2]

   China’s National Computer Virus Emergency Response Center and the National Engineering Laboratory for Computer Virus Prevention issued a warning on 22 April 2026 that multiple counterfeit skill packages containing malicious code had been found in OpenClaw (also called “lobster”) agent skill repositories; once installed, calling those skills silently downloads a trojan in the background, steals private information and may cause financial loss. The center advises auditing prompts, external links and scripts before installation, isolating the agent from production and everyday systems or applying strict access control, and using file hashes to look up and submit suspicious installers, executables or skill packages on the national virus analysis platform (virus.cverc.org.cn). This warning predates the later CNVD notice of 11 May and is an earlier official alert of the same event class.

Vendor guidance

1. 1
   Remove suspicious Skills mitigate^[1]

   Check the advisory list, remove suspicious skills and revoke permissions.