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

Two critical vulnerabilities in Paperclip, an AI agent control plane, allow attackers to execute commands on a server or a developer's computer. The first vulnerability (CVE-2026-41679) requires no prior account or interaction, enabling attackers to import malicious agents and run commands with server privileges. The second vulnerability (GHSA-xfqj-r5qw-8g4j) stems from a DNS rebinding attack in local mode, allowing attackers to bypass authentication and install agents. Both vulnerabilities have been addressed in version 2026.416.0, which requires an upgrade. Rapid7 has released a Metasploit module to automate the attack chain, and CISA has classified the flaw as automatable. The vulnerabilities stem from a lack of proper authorization checks and trust in network locations, potentially exposing sensitive data and internal services.