Breaking

OpenAI Says Unreleased Model Produced Working Root Exploits for 948 Linux Kernel Flaws

November 5, 2026 at 06:50 EST

Security
Foundation Models

OpenAI says a single run of one of its unreleased AI models produced working root exploits for 948 flaws in the Linux kernel, at a total cost of $1.2 million, or about $1,279 per flaw.

OpenAI · AI-driven offensive security

One AI run reportedly produced root exploits for 948 Linux kernel flaws

OpenAI says an unreleased model turned 948 kernel vulnerabilities into working root exploits for about $1.2 million, roughly $1,279 per flaw. The company's threat model: a low-privileged account on an unpatched system is effectively root.

948

Linux kernel flaws with working root exploits in a single run

$1.2M

Total cost of the run

~$1,279

Implied cost per exploited flaw

Scale of the run, drawn to scale

Two separate scales: each block below shows its own unit, so the two columns are not compared directly.

948 flaws

1 block = 50 flaws (≈19 blocks)

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.