OpenAI Launches GPT-6 Cyber, Uncovers 81 Firefox Flaws and 948 Linux Exploits

OpenAI has introduced GPT-6 Cyber, a new artificial intelligence model designed to support cybersecurity experts in identifying software vulnerabilities, testing potential security threats and developing solutions. The company says the model has already delivered significant results, including uncovering 81 previously unknown high-severity vulnerabilities in Firefox and creating working exploits for 948 Linux kernel vulnerabilities.

According to OpenAI, GPT-6 Cyber represents a major improvement over its predecessor, GPT-5.6 Cyber. The latest model offers better vulnerability detection, improved exploit development and greater accuracy when conducting security assessments. It is also designed to follow authorised testing boundaries more consistently, reducing the chances of investigating or attacking systems outside the approved scope.

GPT-6 Cyber delivers improved cybersecurity performance

OpenAI highlighted several improvements in the model's performance across cybersecurity benchmarks. On ExploitGym, a test that evaluates an AI model's ability to exploit specific vulnerabilities in remote systems, GPT-6 Cyber achieved a score of 48.5%. Its predecessor, GPT-5.6 Cyber, scored 32.1%, representing a 51% relative improvement.

The model also showed stronger results in web penetration testing, recording 40.3% vulnerability recall compared with 32 per cent for GPT-6 Astra.

OpenAI further tested whether the AI could respect security testing restrictions. Without additional safety prompts, GPT-6 Cyber avoided exploiting services outside the authorised scope in 99.93 per cent of cases. When safety prompts were enabled, the success rate reached 100 per cent.

Based on these capabilities, OpenAI has assigned GPT-6 Cyber a “Critical” cybersecurity classification under its Preparedness Framework, higher than the “High” rating given to GPT-5.6 Cyber.

Firefox and Linux vulnerabilities identified

The model's security research has already produced notable findings. OpenAI revealed that GPT-6 Cyber successfully reproduced 1,724 Linux kernel vulnerabilities and developed working exploits for 948 of them.

The company has shared 304 critical findings, along with candidate patches, with Linux maintainers. Efforts to address the remaining vulnerabilities are continuing.

In Firefox, the model identified 81 previously unknown high-severity security flaws. OpenAI provided Mozilla with reproducible tests and proposed fixes, with the browser developer having already released patches for several issues. Researchers also discovered that 30 previously reported Firefox vulnerabilities were more serious than initially believed.

Security teams at companies including Cisco, Calif and CodeRabbit are also using the model for vulnerability research and code reviews. OpenAI has deployed it internally to identify and address weaknesses in its own infrastructure.

GPT-6 Cyber availability and security measures

OpenAI is initially offering GPT-6 Cyber through its API and Codex to a limited group of approved Daybreak Red customers. Wider access is expected to follow.

The company is also introducing safeguards to prevent misuse, including monitoring for unauthorised cyber activities and restrictions on high-risk assistance involving critical infrastructure, such as power plants and water treatment facilities.

Organisations and independent cybersecurity researchers can apply to OpenAI's Daybreak programme to access its security tools and resources.

ABOUT THE AUTHOR

Kahekashan

[email protected]

Kahekashan is a passionate technophile with a keen eye for cutting-edge gadgets, emerging technologies, and everything in the digital realm. Raised in a Defence family with strong values and a background in literature, she has consistently pursued excellence in every endeavour. Her last full-time assignment involved content writing with the Indian School of Business.

Read More