Aside from not using passwords or using 2FA, sandboxing helps.
A VM with GPU passthrough set up would be one example (although this is usually a pain to set up and I expect most people aren't doing it).
As a more user-friendly example, if you install an iOS app (local-model LLM and image generation apps exist), the sandboxing provided by the OS ought to be more than enough to prevent keyloggers, short of 0day exploits.
Not as secure as VMs but GPU passthrough with Docker/Podman is much easier to set up, and you can even use the GPU on the host machine at the same time.
Nvidia provides a toolkit to do this [1], getting a GPU into a container is as easy as running `podman run --device nvidia.com/gpu=all`. The process is similar for Docker, but rootless Docker requires some extra steps IIRC.
Comments
Aside from not using passwords or using 2FA, sandboxing helps.
A VM with GPU passthrough set up would be one example (although this is usually a pain to set up and I expect most people aren't doing it).
As a more user-friendly example, if you install an iOS app (local-model LLM and image generation apps exist), the sandboxing provided by the OS ought to be more than enough to prevent keyloggers, short of 0day exploits.
Not as secure as VMs but GPU passthrough with Docker/Podman is much easier to set up, and you can even use the GPU on the host machine at the same time.
Are you giving it access to /dev/dri, or doing some fancier sandboxing?
(Would you even need anything fancier? I think /dev/dri is supposed to isolate users.)
Nvidia provides a toolkit to do this [1], getting a GPU into a container is as easy as running `podman run --device nvidia.com/gpu=all`. The process is similar for Docker, but rootless Docker requires some extra steps IIRC.
[1] https://docs.nvidia.com/datacenter/cloud-native/container-to...