This looks like a neat project that could be useful as part of an AI security stack in particular.
By sandboxing binaries with an allow list of syscalls, you greatly reduce the attack surface for untrusted code/commands.
1 Like
I was just looking for something like this. Was even considering digging into Ansible to build a custom sandboxed VM but this looks like it might do nicely.
Another similar tool that’s actually used in Claude Cowork’s VM for sandboxing commands:
1 Like