So could someone, say, cajole the AI to output a fuzzer with the goal of finding a path to executing arbitrary code? If so, then any similar user-input powered

So could someone, say, cajole the AI to output a fuzzer with the goal of finding a path to executing arbitrary code? If so, then any similar user-input powered AI applications might run this kind of risk.