OpenAI Launches Cybersecurity Model GPT-6 Cyber

According to TechFlow, on November 6, The Verge reported that OpenAI launched the cybersecurity model GPT-6 Cyber. The company said that in tests simulating the earlier AI agent intrusion into Hugging Face, the new model's rate of out-of-scope actions was less than 1%, lower than the approximately 5% for GPT-5.6 Cyber.

In the ExploitGym benchmark measuring models' ability to exploit vulnerabilities, GPT-6 Cyber scored 48.5%, higher than GPT-5.6 Cyber's 32.1% and GPT-6 Astra's 42.4%. OpenAI said the model helped discover 81 previously unknown vulnerabilities in Firefox.

OpenAI has classified the new model's cybersecurity capabilities as "Critical" and added safety measures to detect and block misuse. The model is being made available to some members of the Daybreak Red program and will be expanded to all customers in the program over the coming weeks.

Source: TechFlow