OpenAI Launches GPT-6 Cyber for Security
OpenAI has released GPT-6 Cyber, a powerful new AI model designed for cybersecurity research, to a select group of vetted customers.
The Cliff News | 6 November 2026
OpenAI has unveiled GPT-6 Cyber, a sophisticated artificial intelligence model engineered to identify and exploit security vulnerabilities. The company announced the model's release to a limited cohort of trusted customers on Thursday, assigning it a “Critical” cybersecurity rating under its safety framework.
The new model is intended to equip security researchers with advanced tools to discover flaws, simulate exploitation scenarios, and implement fixes before malicious actors can leverage them. This initiative comes amidst a noticeable rise in attempts to misuse AI technologies by cybercriminals and coordinated groups aiming to replicate or compromise AI models.
Significant Vulnerability Discoveries
OpenAI highlighted the model's effectiveness by revealing that its researchers utilized GPT-6 Cyber to uncover 81 previously unknown vulnerabilities within the Firefox browser. Mozilla, the developer of Firefox, has since classified these findings as high severity and has commenced work on issuing patches.
Furthermore, the AI model played a crucial role in developing working exploits for 948 Linux kernel vulnerabilities. Of these, a significant 588 were identified as novel discoveries. OpenAI has proactively disclosed these critical Linux findings to maintainers, providing proposed patches to address them.
Elevated Cybersecurity Classification
The company has significantly upgraded the cybersecurity classification for GPT-6 Cyber to “Critical” within its safety protocol. This marks an increase from the “High” rating assigned to its predecessor, GPT-5.6 Cyber, underscoring the enhanced capabilities and potential impact of the new model.
OpenAI reiterated its observation of escalating attempts to exploit AI for malicious purposes, including sophisticated attacks orchestrated by cyber actors and groups focused on reverse-engineering AI models. The elevated rating reflects the company's heightened awareness and proactive measures in managing such risks.
Controlled Access and Expansion
Access to GPT-6 Cyber is currently restricted to select customers enrolled in Daybreak Red, OpenAI’s specialized program for advanced and authorized security operations. This tier caters to activities such as penetration testing and exploit development.
OpenAI plans to broaden access to GPT-6 Cyber for all approved Daybreak Red customers over the next few weeks. Concurrently, the company is also making its earlier model, GPT-5.6 Cyber, available to all customers within the Daybreak Red program, effective immediately.