Hacker News new | past | comments | ask | show | jobs | submit
It seems pretty obvious that the solution is auto mode (running a classifier on each action) + sandboxing