Paperclip AI Platform Patches Critical Remote Code Execution

A critical security vulnerability in the open-source Paperclip AI agent orchestration platform allowed attackers to execute arbitrary commands on exposed servers or developers' local machines. The flaw, tracked as CVE-2026-41679 and rated with the maximum CVSS score of 10.0, was fixed in version 2026.416.0, released on August -6, 2026.

The vulnerability affected network-accessible Paperclip deployments running in authenticated mode with default registration settings. According to security researchers, an unauthenticated attacker could register a new account without email verification, sign in, and then create and self-approve a command-line interface (CLI) authorization challenge. This process granted the attacker a persistent board-level API credential.

While direct creation of a new company within Paperclip required instance-administrator privileges, a separate company import path only enforced board-level access. An attacker could exploit this authorization discrepancy by using their board key to call the import API. By supplying a crafted .paperclip.yaml file bundle containing a malicious agent definition that used Paperclip's built-in process adapter, the attacker could...