OpenAI launches GPT-6 Cyber, rates it critical on cyber capability

OpenAI launched GPT-6 Cyber, a version of the GPT-6 Astra model specialised in cybersecurity, initially making it available to approved Daybreak Red customers through the API and the Codex tool.

The company said the model discovered 81 previously unknown high-severity vulnerabilities in the Firefox browser, and built exploits for 948 bugs in the Linux kernel.

It explained that a single AI-supported security workflow identified 2,308 potential vulnerabilities in the Linux kernel and confirmed that 948 of them were critical, at a cost of 1,213,430 dollars in API usage, or about 1,279 dollars per confirmed exploit.

OpenAI classified the model as critical under its preparedness framework, and said access would expand to all Daybreak Red customers soon, while stressing that artificial intelligence did not modify the Linux kernel on its own.

What do these terms mean?

Cybersecurity: Protecting systems, networks and data from hacking and digital attacks.

Preparedness framework: A framework OpenAI uses to assess the capabilities of its dangerous models and classify them before release.

Exploit: A practical method an attacker uses to take advantage of a security flaw in a program.

API: A means that lets other programs connect to and use a service or model.

Linux kernel: The core part of the Linux system that manages hardware and processes on servers and devices.