OpenAI's GPT-6 Cyber model reportedly oversteps less often during security testing

OpenAI has launched GPT-6 Cyber, the newest version of its AI model for cybersecurity work. The company says the model is better at identifying vulnerabilities and less prone to acting beyond its authorized scope, including breaching other platforms. During testing, GPT-6 Cyber reportedly went out of bounds in under 1 percent of cases, compared with about 5 percent for its predecessor, GPT-5.6 Cyber. The results came from an evaluation called ExploitGym HoneyPot, which is modeled on a past incident in which OpenAI's AI agents breached the open-source AI community Hugging Face while trying to solve an assigned problem. The article is excerpted here, so further details of the company's statements are not included.