OpenAI autonomous pipeline: misattributed root shells in a 948-LPE Linux kernel run
- AI company and model
- OpenAI
- Reported
- 5 Nov 2026
- Severity
- Negligible
- Tools it had
- virtual machines, KASAN
- An autonomous OpenAI pipeline searched the Linux kernel for local privilege escalation flaws. Its triage agents built proof-of-concept exploits and ran them in virtual machines with KASAN.
- In an early run, 20 of 200 reported root shells came from a different 0-day than the one assigned. This was a quality-control failure in attributing exploits to flaws.
- Scout agents selected 5,034 audit targets, about 7% of the kernel. The run yielded 948 distinct root causes, each with a patch.
Source: Cyber Press
What was done
OpenAI added an independent check that each exploit uses its assigned flaw. After 10 days, 61 LPEs were patched upstream and 43 had staging patches.
Monitoring added (Developer or vendor) · Patched or fixed (Developer or vendor) · Patched or fixed (Other)
Fix reported: “OpenAI added an independent check that each exploit uses its assigned flaw.”
Risks Silent wrong state or drift (6 incidents)
No source states: AI agent used, money lost or records, company using the AI, sector.
Similar incidents
Evidence
Sources (1)
- OpenAI's AI Model Discovers 948 Linux Kernel Vulnerabilities That Allow Attackers to Gain Root Access · Cyber Press · 2026-11-05
Show every field, with its source and grade
The AI system
Developer
OpenAI
D inferred by us from the described behaviour, reason shown
Tools it was given
virtual machines, KASAN
D inferred by us from the described behaviour, reason shown
What it was for
Discover, validate, exploit and patch local privilege escalation flaws in the Linux kernel
D inferred by us from the described behaviour, reason shown
What happened
Earliest source we hold
2026-11-05
D inferred by us from the described behaviour, reason shown
What was done
OpenAI added an independent check that each exploit uses its assigned flaw.
C one secondary source states it
What went wrong
other
D inferred by us from the described behaviour, reason shown
AI agent
an AI agent acted
D inferred by us from the described behaviour, reason shown
Impact
Severity
negligible
D inferred by us from the described behaviour, reason shown
Why this severity
Nothing beyond the system itself was affected and nothing was left changed. The only failure was misattributed root shells in an early run, corrected with an added check; no effect on others is stated.
D inferred by us from the described behaviour, reason shown
Notes on gaps
No third party harm is stated.
LR-2026-0861 · version 2 · published 2026-11-07 after automatic checks; not yet checked field by field by a person