On March 31, according to Guoshi Direct, it was reported by 360 Digital Security Group that its independently developed 360 multi-agent collaborative vulnerability mining system has discovered a high-risk vulnerability in the OpenClaw platform—specifically, a local file exposure vulnerability due to MEDIA protocol prompt injection bypassing tool permissions. This vulnerability has been officially confirmed by the National Information Security Vulnerability Database (CNNVD) and affects over 50 countries and regions worldwide, with more than 170,000 publicly accessible OpenClaw instances facing security risks. The core risk of this vulnerability lies in the fact that the MEDIA protocol operates at the output post-processing layer, allowing it to completely bypass platform tool policy controls. Even if agents disable all tool calls, attackers can initiate attacks solely based on basic member permissions in group chats, directly stealing sensitive information from servers, which could easily lead to subsequent cyberattacks.