GPT-6 Cyber: OpenAI Uncovers 81 Critical Firefox Vulnerabilities and Develops 948 Linux Exploits

Digital GovernanceGovernance

OpenAI has launched GPT-6 Cyber, a specialised artificial intelligence model designed to identify software vulnerabilities, validate potential cyberattacks and help security researchers develop solutions before malicious actors can exploit system weaknesses. Released on November 5, 2026, the model marks a significant advancement in AI-driven cybersecurity with OpenAI assigning it a ‘Critical’ cybersecurity classification under its Preparedness Framework.

The model has already delivered significant results in real-world security research. OpenAI researchers used GPT-6 Cyber to identify 81 previously unknown vulnerabilities in Mozilla Firefox, all classified as high severity by Mozilla. The findings were shared with Mozilla alongside proposed patches and the browser developer has already started releasing security fixes. The model also identified 30 previously known issues whose severity was higher than initially assessed.

In another major development, GPT-6 Cyber reproduced 1,724 Linux kernel vulnerabilities and developed working exploits for 948 of them, including 588 previously unidentified vulnerabilities. OpenAI has also disclosed 304 critical security findings to Linux maintainers along with proposed patches to support remediation.

The model recorded a 48.5% success rate on the ExploitGym cybersecurity benchmark, compared with 32.1% for its predecessor, GPT-5.6 Cyber, representing a 51% relative improvement. It also demonstrated stronger capabilities in identifying vulnerabilities in web applications without access to their source code.

Strategic Cybersecurity Impact

The launch highlights the growing importance of AI-powered vulnerability detection in protecting digital infrastructure, enterprise networks and critical software systems. Linux supports a substantial portion of global computing infrastructure while Firefox is widely used for internet access. Faster identification and remediation of vulnerabilities could strengthen the security of these digital ecosystems.

However, advanced exploit-development capabilities also introduce potential misuse risks. OpenAI has therefore restricted initial access to selected, verified cybersecurity professionals through its Daybreak Red programme which supports authorised penetration testing and security research.

The development signals a shift towards AI-assisted cybersecurity, where identifying vulnerabilities, validating threats and developing security patches can become faster and more scalable. The broader challenge will be ensuring that defensive capabilities advance faster than malicious cyber threats.