Unrestricted System Access: The Security Crisis Emerging from AI Coding Agent Deployments

With 220,000+ AI coding agents exposed on public servers and developers deploying them locally on Mac Minis, these systems operate with unchecked root-level access to files, credentials, and networks. Two critical CVEs (CVSS 8.8 and 9.6) enable unauthenticated remote code execution in cloud deployments, while malicious skill packages pose equally severe risks to local machines. The architectural flaw creates a unified threat surface across deployment models—making local deployments as vulnerable as exposed cloud servers without proper sandboxing.