Back to News
Technology
Nov 5, 2026101 views2 min read

OpenAI AI Agents Find 948 Linux Security Vulnerabilities in Single Automated Run

OpenAI built an autonomous security pipeline that identified 948 working exploits for Linux privilege-escalation vulnerabilities in a single run costing about $1.2 million in API credits. The system analyzed 2,308 potential flaws and successfully reproduced 1,724 of them. Linux maintainers had already patched 61 of the identified vulnerabilities by the time the research concluded.

OpenAI AI Agents Find 948 Linux Security Vulnerabilities in Single Automated Run
Source:The Neuron

OpenAI has demonstrated an autonomous security pipeline that identified and developed working exploits for 948 distinct Linux privilege-escalation vulnerabilities in a single automated run, the company announced November 5. The experiment cost approximately $1.21 million in API credits, or about $1,279 per distinct vulnerability with a working exploit.

The system analyzed 2,308 potential security flaws in the Linux kernel and successfully reproduced 1,724 of them. Its agents ultimately developed working exploits for 948 of those, all of which could allow someone with limited access to a machine to gain full administrator privileges.

The scale of the experiment is striking compared to earlier work. In July, a system powered by GPT-5.6 Sol found and exploited 166 distinct vulnerabilities. The newer system produced 948 in its October run.

Linux maintainers had already patched 61 of the identified vulnerabilities upstream by the time OpenAI's research concluded. Another 43 had patches in preparation, while 299 were being publicly reported or discussed.

OpenAI said the attacks were developed in a controlled research environment and required an existing foothold on the target computer. The proposed fixes also need expert review before deployment.

The experiment demonstrates how AI can dramatically increase the volume of security vulnerabilities discovered in widely used software. Linux maintainers now face the challenge of assessing those findings quickly enough to get verified fixes into users' hands.

The research comes as OpenAI also announced plans to acquire Pierre Computer, a developer-tools company whose code-navigation technology powers parts of Codex. The deal advances OpenAI's ambitions to compete more directly with GitHub as AI agents take on more software development work.