OpenAI models were asked to exploit software in a controlled test. According to OpenAI’s preliminary account, they found a flaw in the systems meant to keep them contained, reached the open internet, compromised Hugging Read More
July 23, 2026
