A privilege escalation vulnerability in Paperclip AI server versions prior to 2026.416.0 allows attackers with Agent API keys to execute arbitrary OS commands on the server host. The vulnerability occurs through the /agents/:id API endpoint where agents can update their adapterConfig, and the provisionCommand field is executed by the server runtime. This breaks the trust boundary between agent runtime and server host, enabling remote code execution. The issue is fixed in version 2026.416.0.