OpenAI Unleashes GPT-6 Cyber — but Only for a Vetted Few

OpenAI has a new cyber specialist, and this one is not getting an open invitation. The company released GPT-6 Cyber on November 5 to a select group of approved security customers, with a cybersecurity capability rating of critical — the highest level in its safety framework.

The mission? Help defenders find software vulnerabilities, test how they could be exploited and develop fixes. But the same skills that make the model useful for security work also make access a very big deal.

The Linux and Firefox findings

OpenAI says its researchers used GPT-6 Cyber to reproduce 1,724 Linux kernel vulnerabilities and develop working exploits for 948 of them. The company says it disclosed 304 critical findings to maintainers with proposed patches.

The model also produced evidence and reproducible tests for 81 previously unknown Firefox vulnerabilities that Mozilla classified as high severity, according to OpenAI. The company says it shared proposed patches with Mozilla, which has begun releasing fixes.

Who gets in?

For now, GPT-6 Cyber is available through the API and Codex to a select group of customers in Daybreak Red, OpenAI’s access tier for advanced, authorized security work. OpenAI says it plans to expand access to all approved Daybreak Red customers.

The company says access comes with identity verification and security requirements, plus safeguards intended to detect misuse and block high-risk activity against critical systems. It also recommends isolated testing environments and human oversight for high-risk work.

As Axios reports, the race is on: Can defenders find and fix weaknesses faster than attackers can exploit them? GPT-6 Cyber is OpenAI’s latest bet on the defense.

Read more about this story