OpenAI releases GPT-6 Cyber — AI autonomously finds and exploits...

OpenAI releases GPT-6 Cyber — AI autonomously finds and exploits vulnerabilities

On November 5, 2026, OpenAI introduced GPT-6 Cyber, a cybersecurity model that independently identifies vulnerabilities and creates working exploits within researcher-defined parameters. It scored 48.5% on the ExploitGym benchmark and received a Critical rating under the Preparedness Framework for the first time.

Compiled by Sergey KostenchukPublished 2026-11-05Updated 2026-11-05

2026-11-05 Safety OpenAI

🛡 OpenAI releases GPT-6 Cyber — AI autonomously finds and exploits vulnerabilities

On November 5, 2026, the cybersecurity model identifies vulnerabilities and creates working exploits within researcher-defined parameters. It achieved a score of 48.5% on the ExploitGym benchmark, compared to 42.4% for the flagship GPT-6 Astra. Using this model, 81 high-severity vulnerabilities were found in Firefox, and 948 exploits were identified in the Linux kernel, 588 of which were previously unknown.

🌍 For the first time, an OpenAI model has received a Critical cyber rating — an acknowledgment that AI can independently find and exploit unknown vulnerabilities in well-protected systems, while the defensive side lags behind.

👤 The model is currently available only to approved Daybreak Red participants via API and Codex, with general client access expected in the coming weeks. It is recommended to update Firefox, as the discovered vulnerabilities are already being patched.

Source 1: https://openai.com/index/introducing-gpt-6-cyber/ Source 2: https://www.theverge.com/ai-artificial-intelligence/1027599/openai-gpt-6-cyber-launch

Sources