CVE-2026-41208: Paperclip AI Agent Privilege Escalation to RCE

The National Vulnerability Database has detailed CVE-2026-41208, a high-severity privilege escalation vulnerability in Paperclip’s AI agent orchestration platform. Specifically, versions of @paperclipai/server prior to 2026.416.0 are affected. This flaw enables an attacker with an Agent API key to execute arbitrary operating system commands on the Paperclip server host, bypassing intended security boundaries.

The vulnerability stems from agents being permitted to update their own adapterConfig via the /agents/:id API endpoint. Critically, the adapterConfig.workspaceStrategy.provisionCommand field is later executed by the server runtime. This allows a malicious or compromised agent to inject arbitrary shell commands, which the Paperclip server then executes during workspace provisioning. This directly leads to remote code execution (RCE) on the server host, breaking the trust model between agent configuration and server execution.

This is a critical flaw with a CVSS score of 8.8 (HIGH). It transforms a seemingly contained agent compromise into a full server takeover. Defenders must recognize that an attacker only needs an agent credential to achieve RCE. The fix is available in @paperclipai/server version 2026.416.0.