Paperclip AI Flaws Let Attackers Run Host Commands via Malicious Agent Imports

Two critical vulnerabilities in Paperclip, an open-source AI agent control plane, allow remote code execution on server or developer machines via malicious agent imports. CVE-2026-41679 enables unauthenticated attackers to register and import a malicious agent configuration that executes commands with server privileges due to insufficient access controls during company creation. A second path exploits DNS rebinding in local_trusted mode, where a malicious website can trigger command execution on localhost by rebinding a hostname to 127.0.0.1 and invoking the import API. A third vulnerability, GHSA-xfqj-r5qw-8g4j, exposes sensitive API routes without proper authentication, potentially leaking control-plane details and allowing unauthorized access to heartbeat data and system health information.

AI-extracted · verify before operational use