OpenAI says its AI built 948 working Linux exploits in one run, for $1.2 million
On 4 November 2026, OpenAI published research on an AI pipeline that, in one mid-October run with an upcoming cyber model, produced 948 working Linux privilege escalation exploits. The run cost $1,213,430 in API credits, about $1,279 per exploit. It ran in a controlled research setting, and human researchers reviewed the results before reporting bugs and patches to Linux maintainers.
The short version
- 01OpenAI says its pipeline reproduced 1,724 Linux kernel vulnerabilities and built 948 working exploits that let an unprivileged local user gain root access.
- 02The whole run cost $1,213,430 in API credits, about $1,279 per exploit, which OpenAI calls an upper bound.
- 03It used an upcoming OpenAI cyber model that has not been released.
- 04Human researchers reviewed the results and are reporting bugs and patches to Linux maintainers.
- 05OpenAI's advice: keep systems patched, because an out-of-date system with a low-privileged account is close to handing over root.
At a glance
| Status | Confirmed (OpenAI's own research post) |
|---|---|
| Published | 4 November 2026 |
| Company | OpenAI |
| Target | The Linux kernel, on a stock desktop system, in a controlled research environment |
| Run date | Mid-October 2026 |
| Result | 948 working local privilege escalation exploits |
| Cost | $1,213,430 in API credits |
| Model | An upcoming OpenAI cyber model, not yet released |
What did OpenAI announce?
OpenAI published a research post called "A thousand fewer paths to root" on 4 November 2026. It describes an automated pipeline that finds, tests and patches security flaws in the Linux kernel, the core of Linux.
In a run in mid-October 2026, using an upcoming cyber model, the pipeline dynamically reproduced 1,724 vulnerabilities. It built working exploits for 948 flaws that let an unprivileged local user gain root access, which means full control of the computer.
OpenAI says the target was a commonly used Linux distribution running a stock desktop operating system, not a weakened one.
How much did it cost?
The run cost $1,213,430 in API credits, which OpenAI works out as $1,279 for each working exploit.
OpenAI says it focused on measuring what the model can do, not on cost, and that better engineering could cut the price at each stage. It calls its figures an upper bound.
Is this a real attack or a test?
It is a test. OpenAI says it ran the pipeline in a controlled research environment. Human researchers reviewed the exploits, patches and severity before anything was reported.
OpenAI says it began with critical bugs, following guidance from the Linux kernel community, and is working with Patch the Planet partners on more reports and patches.
What does OpenAI warn about?
OpenAI says that over 100 flaws of this kind were patched, or had patches queued by other contributors, within 10 days while its scan was running.
…having access to a low-privileged account on an operating system, that isn’t up-to-date on its latest available patches should be considered functionally equivalent to root access for a well-resourced threat actor.
OpenAI, in its research post
Why does it matter?
Security work that once needed a separate investigation for each flaw can now be run as one process, and OpenAI puts a price on it: about $1,279 per working exploit.
For ordinary people the advice is simple: install updates. OpenAI says unpatched systems are the weak point.
What's next?
OpenAI describes the model as upcoming, so there is no release date to report. It says maintainers are moving quickly to patch the bugs, and that more reports are coming.
Questions people ask
Did OpenAI hack real computers?
No. OpenAI says it ran the pipeline in a controlled research environment against a stock desktop system.
What is a privilege escalation exploit?
A trick that lets someone who already has a normal account on a computer gain full control, known as root access.
How much did the run cost?
$1,213,430 in API credits, about $1,279 per working exploit, according to OpenAI.
Which model did OpenAI use?
An upcoming cyber model that OpenAI has not released.
What should I do?
Keep your devices and software updated. OpenAI says systems that are not up to date are the weak point.
Is this confirmed?
Yes. It comes from OpenAI's own research post of 4 November 2026.
Sources
Why "Confirmed"? We label a story "Confirmed" when the company or government announced it itself. Spotted an error? Email hello@openloopnews.com and we will correct it.
