Share
OpenAI has unveiled an autonomous system that hunts for flaws in the Linux kernel, and its first run produced working root exploits for 948 distinct vulnerabilities.
The total API bill came to $1,213,430. That works out to an average of $1,279 per exploit.
What the pipeline actually found
OpenAI announced the pipeline on November 4, 2026. Its first evaluation ran in mid-October 2026 against the Linux kernel master branch.
The system examined approximately 2.8 million lines of security-relevant code. That slice represents approximately 7% of the entire kernel.
From that code, the pipeline flagged 2,308 candidate vulnerabilities in a single run.
It then tried to confirm each one in isolated testing environments. It successfully reproduced 1,724 of them.
For 291 cases involving information leaks, it generated proofs-of-concept.
The headline number is the 948 functioning local privilege escalation exploits, known as LPEs. An LPE lets someone who already has ordinary access to a machine promote themselves to root, the account that controls everything.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
We respect your privacy. Unsubscribe anytime.
According to the research findings, these exploits worked even on standard desktop Linux setups running typical security mitigations.
The patching gap
As of the announcement, 61 of the vulnerabilities had been fixed upstream. Another 43 were moving through a staged patch process.
Add those together and you get 104 bugs either fixed or in progress. Against 948 working exploits, that leaves a large backlog still waiting on human review.
How we got here
This is not OpenAI’s first pass at the kernel. In June 2026, the company published its “Patch the Planet” disclosures.
Those disclosures highlighted 24 LPE exploits found by an earlier model, GPT-5.5 Cyber. Going from 24 to 948 in roughly five months is a jump of nearly 40 times in confirmed root exploits.
The comparison is not perfectly apples to apples. The new figure comes from a dedicated autonomous pipeline, while the earlier count came from a model, and the scope of each effort may differ.
What this means
For defenders, the most important figure may be the $1,279 average per exploit. Kernel-level root exploits have historically required skilled specialists and significant time to develop.
Local privilege escalation bugs require an attacker to already have some foothold on a machine. That is a meaningful barrier, though attackers often chain an LPE with a separate entry point to take full control.
The pipeline examined approximately 7% of the kernel. If that portion produced 2,308 candidates, the remaining code is presumably not bug-free, and future runs could widen the net.
