OpenAI’s A.I. Tried Breaching 4 Other Targets, Without Prompting