Comment by MikeBattaglia

> Exploits will always be able to chained, and everything will always be exploitable. Including AI systems.

Good point. Also, discovering exploits is also equivalent to the halting problem. So both things are impossible in the general case.