Sandboxing

#1

Post by minimaxir » Wed, Aug 05, 2026, 5:26 PM UTC

Re: Sandboxing

#2

Post by bearjaws » Wed, Aug 05, 2026, 5:29 PM UTC

This really isn't enough IMO.

I recently made a bash script to run in a local directory, boot up a lima VM, setup a tailscale network (I like to access from phone + laptop), and completely segment off everything.

Honestly even with this new setup, I am still paranoid, but its a lot better than what I did before (claude code running on my Mac)

Re: Sandboxing

#3

Post by kiran-ravi » Wed, Aug 05, 2026, 5:47 PM UTC

There are different levels of sandboxing including kernel level for highest protection. definitely depends on the usecases. industry is moving towards on-demand cloud sandboxes.

Re: Sandboxing

#4

Post by no-name-here » Thu, Aug 06, 2026, 2:05 AM UTC

bearjaws wrote: post #2

This really isn't enough IMO. I recently made a bash script to run in a local directory, boot up a lima VM, setup a tailscale network (I like to access from phone + laptop), and completely segment off everything. Honestly even with this new setup, I am still paranoid, but its a lot better than what I did before (claude code running on my Mac)

> I recently made …

It sounds interesting - is the code available somewhere?

Re: Sandboxing

#5

Post by brof » Fri, Aug 07, 2026, 4:34 PM UTC

> On macOS, Zed uses Apple's Seatbelt sandbox through sandbox-exec.

From `man sandbox-exec`, "The sandbox-exec command is DEPRECATED." However, a lot of AI harnesses seem to use it anyway - I wonder what this says about long-term Apple support.