Critical Ruflo Vulnerability Allows Full Control of AI Agent Platforms

Noma Labs has disclosed a critical vulnerability (CVE-2026-59726) in the Ruflo AI hosting platform, allowing unauthenticated attackers to execute arbitrary commands and take full control of AI agent environments. The flaw, known as 'RufRoot', has a maximum CVSS score of 10.0 and enables attackers to exploit a Model Context Protocol (MCP) bridge that is open by default. This vulnerability allows attackers to steal sensitive API keys, read user conversations, and tamper with the AI's memory, influencing future interactions. The vulnerability was responsibly disclosed on June 30, 2026, and Ruflo released a patch within 24 hours, locking down the platform's configuration. However, experts warn that simply patching the software does not guarantee safety, as compromised agents may still operate. Organizations are advised to rotate credentials and audit AI memory for tampering. The incident highlights the emerging risks associated with AI systems.

Key Points: • The Ruflo platform vulnerability allows full remote control with a single unauthenticated request. • Attackers can tamper with AI memory, influencing future responses even after the attack ends. • Organizations must audit AI memory and rotate credentials post-exploitation to mitigate risks.