🔴 Critical | Source: The Hacker News
A critical vulnerability (CVE-2026-59726, CVSS 10.0) in Ruflo, an open-source AI agent harness for Claude Code and OpenAI Codex, allows unauthenticated attackers to execute arbitrary commands remotely. The flaw also enables AI memory poisoning via the Model Context Protocol (MCP), meaning attackers could manipulate the AI agent’s context and behaviour. All versions prior to 3.16.3 are affected, making immediate patching essential for any team running Ruflo in their AI development pipelines.
Security Architect’s Take: Audit your AI development and automation pipelines immediately for any use of Ruflo and upgrade to version 3.16.3 or later without delay. Given the CVSS 10.0 rating and unauthenticated RCE vector, treat any exposed Ruflo instance as potentially compromised and review associated AI agent logs for signs of memory or context tampering.
Original advisory: Ruflo MCP Flaw Lets Unauthenticated Attackers Run Commands and Poison AI Memory