AI Agent Sandboxes Stop Escapes. They Don't Tell You What Happened Inside

Docker has introduced Docker Sandboxes, which run AI coding agents inside isolated microVMs with their own kernels and controlled network access, enhancing security compared to traditional Docker containers. This approach prevents agents from escaping to the host system but does not provide visibility into the agents' internal actions during execution.