Security for local AI and tools like Nono

Questions
1

Hi all,
I’ve been looking at running some local AI but am concerned about the security implications I’ve seen a few tools like Nono GitHub - nolabs-ai/nono: agent runtime security - zero trust, zero setup, zero latency. · GitHub which looks to be working to address this but I’m still not too sure how concerned I should be about localAI and/or what the best approach to securing it is.

Are there any good guides on security for local AI and/or what tech is recommended for locking it down.

Thanks all

2

Thanks, that looks interesting! I’ll definetly take a closer look at it.

I’m also just starting to learn about it, so I can’t give any “best practices”. However, just yesterday I saw a simple overview in the Pi docs that mention a few alternatives: Run Pi in an isolated environment and Run Pi safely.

So far my approach was to put the agent in a rootless podman container, give it a bind mount to pwd and connect it to a local inference engine. However, I wanted to take a closer look at container sandboxes (podman issue), OpenShell and now also nono. If I understood it correctly you can combine nono with your own container isolation as it’s just a single binary, right?