OpenAI launches GPT-6 Cyber to automate vulnerability detection
New AI model shows significant gains in identifying software flaws, but questions linger about its readiness for widespread use.
The OpenAI logo is displayed on a smartphone screen held in one hand, with the Daybreak visual in the background (Samuel Boivin/NurPhoto/Getty Images)
OpenAI released a new artificial intelligence model on Nov. 5, 2026, specifically designed to assist in cybersecurity tasks. Known as GPT-6 Cyber, the system is intended to help identify, validate, and propose fixes for software vulnerabilities.
In testing, the model achieved a 51% relative improvement on ExploitGym benchmarks compared to its predecessor, GPT-5.6 Cyber. The system successfully identified 81 previously unknown high-severity vulnerabilities in Firefox and developed exploits for 948 Linux kernel vulnerabilities, including 588 bugs that were previously undocumented.
OpenAI has classified the model as Critical under its internal Preparedness Framework. The company stated that the system includes safeguards designed to block high-risk activities, such as attacks on critical infrastructure, and reported that the model avoided exploiting out-of-scope services in 99.93% of tests.
The release comes as AI laboratories face increased scrutiny regarding their role in protecting digital infrastructure. The industry is currently shifting its focus from the process of finding vulnerabilities to the labor-intensive task of patching them.
Max Justicz, CTO and co-founder of Charge Robotics, noted the efficiency of using AI for vulnerability research. He stated that he previously used ChatGPT to find bugs in a body of code by referencing past issues, resulting in the discovery of complex vulnerabilities within two hours.
Access to the new model is currently limited to Daybreak Red members via API and Codex. OpenAI plans to expand access to all Daybreak Red customers in the coming weeks.