OpenClaw 2.0 ships, and its maintainers explain how they secured it

OpenClaw, a personal AI assistant that runs on users' devices, illustrates why restricting an agent to a folder requires checking links inside it. In a GitHub interview, a maintainer describes blocking symbolic links that led outside a configured workspace. These links let a location inside one folder point to another folder outside it.

The maintainer says this broke many existing setups and drew many complaints, because users had been linking other folders into their workspace. They also say that leaving the gap open would have invited a flood of security reports. This is the speaker's own account, not an independent security assessment, but it shows a concrete cost of enforcing a workspace restriction: users lose convenient access to linked folders beyond that boundary, in exchange for a safer default.