News2 min read

OpenAI's GPT-6 Cyber built working exploits for 948 Linux flaws

OpenAI released GPT-6 Cyber to approved security teams. Its researchers used it to reproduce 1,724 Linux kernel flaws and exploit 948 of them.

The words GPT-6 Cyber under a warm sunrise over a dark, curved horizon, with faint digits scattered across the sky
Image: OpenAI

OpenAI released GPT-6 Cyber on Nov 5, 2026, a model trained to find security flaws, prove they can be exploited and help write fixes (OpenAI). It is not public. It is available through the API and Codex to a select group of approved customers of Daybreak Red, OpenAI’s program for authorized security testing (OpenAI).

Linux kernel flaws reproduced
1,724
with working exploits
948
critical findings disclosed to maintainers
304
Source: OpenAI

What happened

OpenAI’s researchers pointed the model at the Linux kernel. Axios reports the 948 exploits gave administrator-level access, cost about $1.2 million in API credits, and were reviewed by human researchers before they were reported (Axios). OpenAI says it disclosed the 304 critical findings to the kernel’s maintainers with candidate patches, following the Linux community’s guidance, and is working on getting the rest fixed (OpenAI).

ExploitGym score
GPT-5.6 Cyber32.1%
GPT-6 Cyber48.5%

Source: OpenAI. ExploitGym tests whether a model can exploit a given vulnerability.

The announcement also lists:

  • Firefox: evidence for 81 previously unknown vulnerabilities that Mozilla classified as high severity. Mozilla has released fixes for many of them (OpenAI).
  • Rating: OpenAI classifies the model as Critical in cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber (OpenAI).
  • Staying in scope: in a test where unrelated services were left easy to attack, the model avoided exploiting them in 99.93% of cases without its deployment safety prompt and 100% with it (OpenAI).
  • Safeguards: monitoring for misuse, and IP allowlisting by default for Daybreak Red API access, so requests come from approved networks, with limited exceptions (OpenAI).

Why it matters

Axios says AI now finds and tests security weaknesses at a scale that could overwhelm companies’ ability to fix them before attackers use them, and that security professionals expect open-source models to match this soon (Axios). OpenAI says there is “a limited window to use frontier models to secure software”, and plans to widen access to organizations of every size, independent researchers and open-source maintainers (OpenAI).

OpenAI says access will expand to all Daybreak Red customers “soon” and gives no date (OpenAI).

Sources

Type to search…