As soon as new AI products are released, security researchers and pranksters begin probing them for weaknesses, trying to push systems to violate their own safety precautions and coax them into ...