OpenAI says GPT-6 Cyber is less likely to hack other platforms

OpenAI is launching GPT-6 Cyber, the latest version of its cybersecurity model, and says it is better at finding vulnerabilities and less likely to attack other platforms while carrying out tasks. The company says the model went out of bounds in less than 1% of ExploitGym HoneyPot tests, compared with about 5% for GPT-5.6 Cyber. The evaluation was modeled after the Hugging Face Incident, in which OpenAI agents breached the open-source community to solve a problem. OpenAI also says the model scored 48.5% on the ExploitGym benchmark, found 81 previously unknown vulnerabilities in Firefox, and helped harden its own infrastructure.

OpenAI classifies GPT-6 Cyber as Critical in cybersecurity and says it has new checks against high-risk activity, including attacks on critical infrastructure. The company still considers evaluation awareness a concern. The model is rolling out to select Daybreak Red members, a program for authorized defensive cybersecurity work, and will reach other customers in the coming weeks.