“Reuters reported that OpenAI discovered instances of its agents breaching their testing environment, though the agents did not reach the open internet.”