Science & Tech · Cybersecurity · 1 hr ago
OpenAI pipeline finds 948 Linux flaws that could give users root access
OpenAI used an automated AI pipeline to find Linux kernel security flaws in a run in mid-October 2026.
The flaws could let someone with an ordinary local account gain root access, which gives broad control over a Linux system.
The pipeline identified 2,308 possible issues, reproduced 1,724 in controlled tests and generated 1,169 working exploits, which analysis linked to 948 distinct vulnerabilities.
It also produced candidate patches, and human security researchers reviewed the findings before they were reported to maintainers.
The run took 30 hours and cost about $1.21 million; researchers said they found at least 10 exploitation techniques they believe were previously undocumented.
OpenAI said it has sent hundreds of reports and patches and is working with partners on more, while the findings show that AI can help defenders fix flaws but could also help attackers find and exploit them.
OpenAI said its automated cybersecurity pipeline identified 948 distinct Linux kernel vulnerabilities that could let local users gain root access.
The system analyzed 2,308 candidate findings and dynamically reproduced 1,724 vulnerabilities.
It generated working exploits and candidate patches, which human security researchers reviewed before reports were sent to maintainers.
OpenAI said the mid-October run using an upcoming cyber model cost about $1.21 million in API credits.
Researchers found at least 10 exploitation techniques they believed had not been documented before.
- Who
- OpenAI developed and ran the cybersecurity pipeline.
- What
- It identified 948 distinct Linux kernel vulnerabilities that could enable local privilege escalation and found at least 10 previously undocumented exploitation techniques.
- When
- The main run took place in mid-October 2026.
- Where
- The vulnerabilities were in the Linux kernel; the system reproduced them in controlled environments.
- Why
- OpenAI said the goal was to help defenders find and fix vulnerabilities before attackers could use them.
This story does not have two clearly opposing sides.
An earlier OpenAI campaign using GPT-5.6 Sol found and exploited 166 unique local privilege escalation vulnerabilities, according to Interesting Engineering.
OpenAI ran its pipeline, which reproduced 1,724 vulnerabilities within 30 hours and identified 948 distinct local privilege escalation vulnerabilities.
OpenAI posted about its automated vulnerability discovery, exploitation and patching pipeline, according to the post reproduced by Interesting Engineering.
- Candidate findings
- 2,308
- Vulnerabilities reproduced
- 1,724
- Distinct local privilege escalation vulnerabilities
- 948
- Information-leak proofs of concept
- 291
- API cost
- $1,213,430
- Exploitation techniques
- At least 10 believed to be previously undocumented
3 sources
What people are saying
Our prototype identified and analyzed 2,308 candidate Linux kernel findings, dynamically reproduced 1,724 vulnerabilities, and produced 948 local privilege escalation (LPE) exploits that could allow an unprivileged local user to gain root access,
The total run cost $1,213,430 in API credits, or $1,279 per LPE exploit.
Posts on X
Our security researchers built an autonomous cyber vulnerability discovery, exploitation, and patching pipeline. In a single run with an upcoming cyber model, it validated and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel at a total API