OpenAI launches GPT-6 Cyber for authorized vulnerability research

OpenAI released GPT-6 Cyber on November 5, 2026, a machine learning model purpose-built for authorized vulnerability research and remediation Reuters. The model is designed to help security researchers identify exploitable vulnerabilities, test how they could be exploited, and develop fixes before attackers do—all within a defined scope of authorized research.

Model Capabilities and Performance

OpenAI classified GPT-6 Cyber as "Critical" for cybersecurity under its Preparedness Framework The Register. The company reported a 51% relative improvement over its predecessor, GPT-5.6 Cyber, on the ExploitGym benchmark—an industry-standard measure of vulnerability-finding performance OpenAI.

The model is described as "more capable at finding and fixing vulnerabilities" and "far more reliable staying within the boundaries researchers set," addressing a key concern in the deployment of powerful cybersecurity tools. OpenAI emphasized that GPT-6 Cyber is engineered with safety constraints to keep usage within ethical and legal frameworks for authorized security work.

Restricted Access and Distribution

Initial access to GPT-6 Cyber is limited to approved Daybreak Red customers, a vetted group of security researchers and organizations The Register. The model is available through OpenAI's API and Codex interfaces, integrating into existing developer workflows. This phased rollout approach reflects OpenAI's strategy of carefully monitoring deployments of high-capability security tools.

The Daybreak Red program represents a deliberate choice to build trust and gather feedback from security professionals before wider distribution. This model allows OpenAI to observe real-world usage patterns, validate safety measures, and refine the system's ability to stay within defined boundaries.

Industry Context

The release arrives amid growing attention to AI's role in both offensive and defensive cybersecurity. By pairing advanced vulnerability discovery with built-in safety constraints, OpenAI is attempting to tilt the scales toward defenders—enabling faster identification and patching of flaws before malicious actors exploit them SecurityWeek.

The 51% performance improvement suggests meaningful progress in the model's ability to handle complex vulnerability scenarios, a critical metric for organizations relying on AI to augment their security teams.