Comment by teaearlgraycold

Can anyone with more experience with systems programming tell me if it’s feasible to whitelist syscalls that are “read only” and allow LLMs free rein as long as their sub-processes don’t mutate anything?