// AI threat desk · 15 Nov 2026
Home/AGENT · Agent Security
Agent SecurityMediumAISECCVSS not assigned

Linux kernel pipeline built 948 working root exploits, OpenAI said

OpenAI said the scan targeted recent code in the kernel’s master branch and tested exploits with common protections enabled.

· 2 min read
AI-generated image of security researchers reviewing kernel code and exploit test results on monitors.
AI-generated image, not a photo of the event.
Key takeaways
  • An autonomous pipeline built working root exploits for 948 unique Linux kernel vulnerabilities in one run, OpenAI said.
  • The run cost $1.2 million in API credits, according to OpenAI.
  • OpenAI said the agents discovered and used at least 10 novel Linux exploitation techniques.

The pipeline scanned the most recent code in the kernel’s master branch. OpenAI said the agents flagged 2,308 potential bugs and confirmed 1,724 of them.

On this page

The pipeline tested exploits against protected desktop Linux systems

The 948 figure represents separate patches created for the bugs behind the exploits, OpenAI said. Its security researchers checked the exploits, fixes and proposed severity before anything went to kernel maintainers.

OpenAI said the agents discovered and used at least 10 novel Linux exploitation techniques, but full technical details were being kept private for safety reasons.

OpenAI compared the scan with earlier kernel testing

OpenAI said the scan cost $1,213,430, or roughly $1,279 for each exploit across the 948 exploits. An earlier scan in late July, powered by GPT-5.6 Sol, yielded 166 unique local privilege escalations at a cost of $513,690.

The company also pointed the pipeline at an adapted version of Google’s kernelCTF. Over 112 hours and about $33,000 in API credits, that test produced 21 local privilege-escalation exploits.

HKCERT listed SUSE Linux Enterprise Server 15 SP7 among affected products

HKCERT said SUSE Linux Enterprise Server 15 SP7 was among the affected products in its listing of SUSE Linux kernel vulnerabilities.

OpenAI said unpatched systems can expose privileged access

OpenAI warned that access to a low-privileged account on an operating system that is not up to date with its latest available patches should be considered functionally equivalent to root access for a well-resourced threat actor.

To contain damage when a flaw is exploited, OpenAI pointed to layered defenses including sandboxing and isolation technologies such as Firecracker and gVisor.

OpenAI said it had already submitted hundreds of reports and patches for critical flaws. It will send fixes for remaining bugs in stages with partners in its Patch the Planet initiative.

OpenAI said more than 100 Linux local privilege escalations had been patched or had staging patches within 10 days while the scan was running.

What to do now

  1. Apply fixes issued by the vendor for the Debian Linux kernel vulnerabilities.
  2. Apply fixes issued by Red Hat for the Red Hat Enterprise Linux kernel vulnerabilities.
  3. Apply fixes issued by SUSE for the SUSE Linux kernel vulnerabilities.

FAQ

What did OpenAI’s Linux kernel pipeline do?

OpenAI said an autonomous pipeline built working root exploits for 948 unique Linux kernel vulnerabilities during a single run.

How many Linux kernel bugs did the agents confirm?

OpenAI said the agents flagged 2,308 potential bugs and confirmed 1,724 of them.

Which Linux products did HKCERT list as affected?

HKCERT listed Debian stable distribution trixie versions before 6.12.113-1 and SUSE Linux Enterprise Server 15 SP7, and also described vulnerabilities in Red Hat Enterprise Linux kernels.

Did the pipeline’s exploits run on protected systems?

OpenAI said the exploits were tested on virtual machines running an unmodified desktop Linux setup with KASLR, SMEP and SMAP enabled.

What is the cost of the OpenAI Linux kernel scan?

OpenAI said the scan cost $1,213,430, or roughly $1,279 for each exploit across the 948 exploits.

How are the remaining Linux kernel fixes being handled?

OpenAI said it would send fixes for remaining bugs in stages with partners in its Patch the Planet initiative.

Sources

  1. Debian Linux Kernel Multiple Vulnerabilities, HKCERTPrimary
  2. RedHat Linux Kernel Multiple Vulnerabilities, HKCERTPrimary
  3. SUSE Linux Kernel Multiple Vulnerabilities, HKCERTPrimary
  4. AI Agents Got Root on Linux via 948 Kernel Flaws in OpenAI Research, SecurityWeek
How we checked this story
ClaimSourceStatus
OpenAI said an autonomous pipeline powered by an unreleased cyber model built working root exploits for 948 unique Linux kernel vulnerabilities during a single run.OpenAI, via SecurityWeekAttributed
OpenAI said the single run cost $1.2 million in API credits.OpenAI, via SecurityWeekAttributed
OpenAI said the mid-October scan targeted the most recent code in the kernel’s master branch.OpenAI, via SecurityWeekAttributed
OpenAI said AI agents flagged 2,308 potential bugs and confirmed 1,724 of them.OpenAI, via SecurityWeekAttributed
OpenAI said the exploits were tested on virtual machines running an unmodified desktop Linux setup with common protections enabled.OpenAI, via SecurityWeekAttributed
OpenAI said the agents produced 1,169 working local privilege-escalation exploits and 291 proofs of concept for information leaks.OpenAI, via SecurityWeekAttributed
OpenAI said the 948 figure represented separate patches created for the bugs behind the exploits.OpenAI, via SecurityWeekAttributed
OpenAI said its security researchers checked the exploits, fixes and proposed severity before anything went to kernel maintainers.OpenAI, via SecurityWeekAttributed
OpenAI said the agents discovered and used at least 10 novel Linux exploitation techniques.OpenAI, via SecurityWeekAttributed
OpenAI said full technical details of the techniques were being kept private for safety reasons.OpenAI, via SecurityWeekAttributed
OpenAI said the scan cost $1,213,430, or roughly $1,279 for each exploit across the 948 exploits.OpenAI, via SecurityWeekAttributed
OpenAI said an earlier late-July scan powered by GPT-5.6 Sol yielded 166 unique local privilege escalations at a cost of $513,690.OpenAI, via SecurityWeekAttributed
OpenAI said the adapted kernelCTF test produced 21 local privilege-escalation exploits over 112 hours and about $33,000 in API credits.OpenAI, via SecurityWeekAttributed
OpenAI said it had already submitted hundreds of reports and patches for critical flaws.OpenAI, via SecurityWeekAttributed
OpenAI said it would send fixes for remaining bugs in stages with partners in its Patch the Planet initiative.OpenAI, via SecurityWeekAttributed
OpenAI said more than 100 Linux local privilege escalations had been patched or had staging patches within 10 days while the scan was running.OpenAI, via SecurityWeekAttributed
OpenAI warned that an unpatched operating system with a low-privileged account should be considered functionally equivalent to root access for a well-resourced threat actor.OpenAI, via SecurityWeekAttributed
OpenAI recommended layered defenses including sandboxing and isolation technologies such as Firecracker and gVisor.OpenAI, via SecurityWeekAttributed
HKCERT said multiple Debian Linux kernel vulnerabilities could enable denial of service, elevation of privilege and sensitive information disclosure.HKCERTConfirmed
HKCERT said Debian stable distribution trixie versions before 6.12.113-1 were affected.HKCERTConfirmed
HKCERT said multiple Red Hat Enterprise Linux kernel vulnerabilities could enable denial of service, privilege elevation, restriction bypass, information disclosure, remote code execution and data manipulation.HKCERTConfirmed
HKCERT said multiple SUSE Linux kernel vulnerabilities could enable denial of service, privilege elevation, data manipulation, restriction bypass, information disclosure and remote code execution.HKCERTConfirmed
HKCERT listed SUSE Linux Enterprise Server 15 SP7 among the affected products.HKCERTConfirmed

Could not verify

  • Whether the 948 vulnerabilities were assigned CVE identifiers is not established
  • Whether any of the generated exploits were used in real-world attacks is not established
  • How many Linux systems were affected by the vulnerabilities is not established
  • Whether the Debian, Red Hat and SUSE advisories describe vulnerabilities from the OpenAI scan is not established
agentsopenai
See all in Agent Security →
Explore with AI
ChatGPT
Claude
Perplexity
Grok