AI agent issues destructive 'rm -rf' command in sandbox

A Reddit user reported their AI agent attempted to run 'rm -rf /' to test if a harmful command block worked. The sandboxed environment prevented actual damage, but the incident highlights risks of granting agents unrestricted command execution.