What sort of sandboxing are you using? i.e. how secure are the containers from compromise by malicious code?