OpenAI Launches GPT-6 Cyber With 51% Performance Improvement To Advance AI-Powered Cybersecurity Research

OpenAI has introduced GPT-6 Cyber, a specialized artificial intelligence model designed to identify, validate, and help remediate software vulnerabilities while operating within explicitly authorized security testing boundaries. The model delivers substantial improvements over its predecessor, GPT-5.6 Cyber, including a 51% relative performance increase on the ExploitGym cybersecurity benchmark.

The November 5, 2026 announcement represents a significant expansion of OpenAI’s efforts to develop advanced AI systems for cybersecurity research, vulnerability discovery, and defensive security operations.

GPT-6 Cyber achieved a 48.5% success rate on ExploitGym, compared with 32.1% for GPT-5.6 Cyber. The benchmark evaluates whether AI models can successfully exploit specified vulnerabilities in controlled environments.

The model also demonstrated improved performance in vulnerability discovery. In OpenAI’s evaluation of web penetration testing without source code access, GPT-6 Cyber identified 40.3% of vulnerabilities, compared with 32% for GPT-6 Astra.

On RemoteExploitBench, a benchmark involving known vulnerabilities in applications including GitLab, PostgreSQL, and Jenkins, GPT-6 Cyber successfully completed 95% of tasks.

Based on these capabilities, OpenAI classified GPT-6 Cyber as Critical in cybersecurity under its Preparedness Framework, compared with the High classification assigned to GPT-5.6 Cyber.

A central component of the model’s development involved strengthening alignment and preventing unauthorized security activities.

In OpenAI’s ExploitGym Honeypot evaluation, GPT-6 Cyber avoided exploiting out-of-scope services in 99.93% of tests without deployment-specific safety instructions, compared with 94.49% for GPT-5.6 Cyber.

With deployment safety instructions enabled, GPT-6 Cyber remained within authorized boundaries in 100% of the evaluated cases.

These improvements are intended to help security professionals use advanced offensive research capabilities without allowing AI agents to expand their activities beyond permitted systems.

OpenAI has also demonstrated the model’s effectiveness in real-world software security research.

During an investigation of the Firefox browser, GPT-6 Cyber identified 81 previously unknown vulnerabilities classified by Mozilla as high severity.

The model also provided evidence showing that another 30 previously identified issues were more severe than initially reported.

OpenAI shared the findings with Mozilla, along with candidate patches, enabling engineers to begin addressing the vulnerabilities.

In separate Linux kernel research, GPT-6 Cyber reproduced 1,724 vulnerabilities and developed working exploits for 948.

That compared with 166 successfully exploited vulnerabilities using GPT-5.6 Sol under the same research framework.

OpenAI disclosed 304 critical Linux findings to maintainers, providing potential fixes as part of its coordinated vulnerability disclosure efforts.

The company also used GPT-6 Cyber alongside other models to investigate vulnerabilities in Brotli, a compression library used by Node.js and other applications.

Beyond open-source projects, security teams at organizations including Cisco, Calif, and CodeRabbit are using the model for vulnerability research and code review.

OpenAI has also incorporated the technology into defensive assessments of its own infrastructure, including investigations of vulnerabilities involving isolated computing environments.

A significant part of the launch focuses on safeguarding critical infrastructure.

OpenAI partnered with industrial cybersecurity company Dragos to evaluate AI models in simulated operational technology environments.

During controlled testing involving a mock power plant, a frontier model identified operational technology assets and discovered two previously unknown vulnerabilities in engineering software used to manage protective relay data.

The findings demonstrated both the defensive value of advanced AI research and the potential risks associated with increasingly capable autonomous cybersecurity systems.

Feedback from Dragos informed OpenAI’s safeguards for industrial control systems, including protections intended to prevent unauthorized activities affecting power plants, water treatment facilities, and other safety-critical environments.

OpenAI is also expanding its Daybreak cybersecurity program, which provides different levels of model access based on security professionals’ needs and authorization.

Daybreak Blue supports verified defenders conducting vulnerability research and validation, while Daybreak Red provides more advanced capabilities for authorized penetration testing and exploit development.

The company has introduced Daybreak Red for Critical Infrastructure, a specialized access category intended for approved organizations protecting industrial and essential service environments.

OpenAI’s $1 billion Daybreak for Frontline Defenders grant program will also subsidize access to advanced security technology for resource-constrained organizations, including government agencies and essential service providers.

Additional safeguards accompanying GPT-6 Cyber include cybersecurity-specific misalignment monitoring, threat detection, stronger credential protections, and enhanced oversight of activities involving critical infrastructure.

OpenAI initially made GPT-6 Cyber available through its API and Codex to a select group of approved Daybreak Red customers, with plans to expand availability.

The company is also broadening access to GPT-5.6 Cyber and GPT-6 Astra through its existing cybersecurity programs.

The launch reflects OpenAI’s strategy of equipping security professionals with increasingly capable AI systems while developing controls intended to reduce misuse.

As AI-generated software becomes more widespread and cyber threats grow more sophisticated, OpenAI is positioning GPT-6 Cyber as a tool for accelerating vulnerability discovery, validating security weaknesses, and helping defenders strengthen critical software infrastructure.

KEY QUOTES:

“GPT‑6 Cyber is a world-class model for cyberdefense. Firefox entered 2026 as one of the world’s most hardened and scrutinized codebases. Even after further hardening with multiple generations of frontier models, GPT‑6 Cyber uncovered previously unknown high-severity vulnerabilities, identified architectural improvements, and helped us ship fixes to millions of users.”

Brian Grinstead, Distinguished Engineer, Firefox

“Dragos put a frontier model on a mock power plant range and told it almost nothing. On its own, it separated the OT assets from the IT noise, then found two zero-days in engineering software that manages protective relay data, exposing protection settings and breaker trip logic. That’s the map an adversary wants before touching anything physical. The skill floor is dropping for the reconnaissance and vulnerability research needed before an attack. Working with OpenAI on this shows them the adversary’s path through an OT network before real adversaries take it, so safeguards get built around how these attacks could unfold, while the models stay useful to the people defending these systems.”

Rob T. Lee, Chief AI Officer, Dragos