OpenAI launches GPT-6 Cyber, access currently limited to approved Daybreak Red customers

OpenAI has launched GPT-6 Cyber, a new cybersecurity model designed to find serious software vulnerabilities and develop working exploits. The company says it has already uncovered hundreds of previously unknown Firefox flaws and developed hundreds of Linux kernel exploits. However, its biggest selling point may be how reliably it stays within the boundaries of authorized security testing.

GPT-6 Cyber finds 81 high-severity Firefox vulnerabilities

OpenAI says GPT-6 Cyber has already uncovered 81 previously unknown high-severity vulnerabilities in Firefox. The company worked with Mozilla, which received reproducible tests and proposed patches for the findings. Mozilla has since started landing fixes for many of these vulnerabilities.

Linux has also received considerable attention from the new model. OpenAI says GPT-6 Cyber reproduced 1,724 vulnerabilities and developed working exploits for 948 Linux kernel vulnerabilities. That is more than five times the 166 exploits produced by GPT-5.6 Cyber’s predecessor research setup.

Interestingly, the model also helped identify 30 existing Firefox issues that were more severe than previously reported. OpenAI says its researchers found 304 critical Linux findings to maintainers, with candidate patches. Other fixes are still in progress.

OpenAI claims a major jump in cybersecurity capabilities

GPT-6 Cyber delivers a 51% relative improvement on ExploitGym compared with GPT-5.6 Cyber. Its score increased from 32.1% to 48.5%. The benchmark measures how well models can exploit a specified vulnerability in a controlled target.

The company also focused heavily on keeping the model within authorized testing boundaries. In its ExploitGym Honeypot evaluation, GPT-6 Cyber avoided exploiting out-of-scope services in 99.93% of tests without deployment prompts. That figure reached 100% when the prompts were enabled.

OpenAI classifies GPT-6 Cyber as Critical in cybersecurity under its Preparedness Framework. GPT-5.6 Cyber previously received a High classification. The stronger capabilities also come with additional safeguards, including misuse monitoring, credential protections, and restrictions around critical infrastructure.

Access remains limited for now

GPT-6 Cyber is initially available through the API and Codex to selected, approved Daybreak Red customers. OpenAI plans to expand access to all Daybreak Red customers soon. The company is also expanding Daybreak Blue access for GPT-6 Astra. Meanwhile, its $1 billion Daybreak for Frontline Defenders program aims to help resource-constrained security teams access advanced cybersecurity tools.