OpenAI says its new GPT-6 Cyber has already uncovered ‘thousands’ of exploits

OpenAI has launched GPT-6, a cybersecurity AI model that delivers “major gains” on key cybersecurity benchmarks over its predecessor, GPT-5.6 Cyber.

In an introductory blog post published on Thursday, OpenAI said that GPT-6 Cyber delivers a 51% improvement over GPT-5.6 Cyber on the ExploitGym cybersecurity benchmark.

GPT-6 Cyber is “already helping to secure real-world software, uncovering and validating thousands of exploits in security-critical codebases such as Linux and Firefox,” said the post.

The news comes days after it was revealed that OpenAI’s upcoming model (now confirmed to be GPT-6 Cyber by OpenAI) had found and patched 948 local privilege-escalation vulnerabilities in Linux. The model was also used to identify 81 previously unknown “high-severity” vulnerabilities in Firefox, “despite extensive prior testing with frontier models.”

If this all sounds a little dangerous to be available to just anyone, you’re right. This is why access to GPT-6 Cyber is part of Daybreak, OpenAI’s cybersecurity program, which is only available to verified individuals and organizations.

In its blog post, however, OpenAI said its goal is to “dramatically” expand Daybreak, and bring “advanced security research to organizations of every size, independent researchers, and open-source maintainers.”

Initially, GPT-6 Cyber is only available to a “select group” of approved Daybreak Red customers (Daybreak access is divided into two tiers: Daybreak Blue is for defensive work, while Daybreak Red is for advanced penetration testing and security research). Access will be progressively expanded to all Daybreak Red customers “soon.”

Want more tech and digital culture news delivered to your inbox daily or sent straight to your device? Sign up for Mashable’s Top Stories newsletter or get Mashable push alerts today.