A critical vulnerability in the open‑source AI hosting platform Ruflo, tracked as CVE-2026-59726, lets unauthenticated attackers seize control of AI agents and run arbitrary code on the underlying system.
The flaw scores a CVSS 10 rating and resides in the Ruflo MCP bridge component, where a single crafted HTTP request can trigger full remote code execution without any authentication as demonstrated by Noma Labs.
Exploiting the vulnerability grants attackers access to stored user conversations, API keys and other credentials, and can corrupt the AI model’s memory to alter its reasoning or inject malicious behaviour that survives a patch according to The Hacker News.
To date there is no evidence of the flaw being exploited in the wild and no threat actor has been linked to the issue, but the proof‑of‑concept shows how a compromised agent could be turned into a rogue admin or used to launch AI‑driven swarms.
Security teams should treat any credentials or tokens stored inside Ruflo instances as compromised, rotate those secrets immediately and rebuild affected containers from verified images.
They should also apply the vendor patch as soon as it becomes available, restrict network access to the MCP bridge to trusted hosts and monitor logs for unexpected outbound connections or sudden changes in agent behaviour.
Organisations that cannot patch right away are advised to disable the MCP bridge feature if it is not required, enforce least‑privilege principles on API keys and consider deploying runtime integrity checks to detect memory tampering.