The vulnerability to potential hackers was found by using a model from OpenAI in one of the most high-profile demonstration of AI's cybercapabilities.