We need more reviewers because finding Linux bugs has become easy, but fixing them is a different challenge.
AI bug scanning can produce many Linux kernel reports, but human maintainers struggle to review them, creating a patch bottleneck and burnout.
Key Takeaways
- Microsoft’s MDASH AI tool found 59 upstream Linux kernel bugs, including one at least 21 years old, showing that AI can discover vulnerabilities faster than humans can review and fix them.
- The main bottleneck is now human review: maintainers are overwhelmed by AI-generated bug reports and patches, and some are limiting their review work.
- Linux’s huge scale makes triage difficult, because a bug may be critical in one deployment but less urgent in another, while Windows is also seeing a high volume of security issues.
- AI may eventually help clear the backlog by generating fixes, but maintainers say the balance between bug discovery and trusted patching has not been reached yet.
- For users and IT teams, the practical takeaway is that more AI-driven reports may make patch cycles look busier without making every issue immediately urgent or quickly resolved.
Microsoft’s FORGE Lab said its AI scanning tool, MDASH, turned up 59 upstream Linux kernel bug reports, including a flaw at least 21 years old. For developers, IT teams, and users of Linux and Windows systems, the finding shows a new security bottleneck: AI can find bugs faster than human maintainers can review, validate, and patch them.
AI made bug discovery cheap, and the flood arrived
In Linux, a bug can become a security bug if someone can abuse it. Linus Torvalds, creator of Linux, said in 2024, “Bugs will happen, and anything can be a security bug if somebody is clever enough to just figure out how to abuse it.”
That helps explain the volume. For 2026, through Nov. 2, the Linux Common Vulnerabilities and Exposures team reported an average of 27 CVEs a day. The scale of Linux also complicates triage: Linux runs smartphones, refrigerators, and supercomputers, so a bug that is dangerous in one deployment may not appear in another.
The pressure is not limited to Linux. In the latest Windows Patch Tuesday security release, Microsoft reported 726 CVEs. The source says the vast majority of bug reports are coming from AI programmers. FORGE Lab said, “Finding bugs and fixing them are drastically out of balance.” Turning its findings into reports took more than four person-months because “human effort [is] the bottleneck.”
Maintainers say the review bottleneck is becoming a crisis
The strain is falling hardest on the people who keep major projects running. On Nov. 10, Lorenzo Stoakes, an ARM engineer and Linux kernel maintainer, said he was giving up reviewing code outside his maintained areas. He had reviewed the memory-management core, misc, reclaim, and Rust-related sections. Stoakes, a kernel contributor since 2014, wrote, “Given the current throughput in the kernel resulting from AI-generated series, I unfortunately lack the bandwidth to generally be able to offer review except in those areas I maintain.”
He said the AI wave had been “horrible,” and was getting worse. At first, he saw low-quality submissions once in a while. Then it was at least once a week, then every few days. “Nowadays it’s pretty much half of what I see, if not more,” he wrote.
Greg Kroah-Hartman, the Linux kernel stable maintainer, gave a blunt assessment at Open Source Summit Europe. “We don’t have enough people to review them, the onslaught of the LLM-generated patches. We need more reviewers. We don’t have enough. All of the maintainers are burnt out, and we have been for ten years,” he said.
AI may also be part of the fix
The flood is not only a problem for AI. The same tools may help clear it. Torvalds said in a recent Linux Kernel Mailing List message that much of the new code in Linux 7.4 rc 2 “was driven by various AI fixes.”
He added that, on the bpf side at least, fixes are now coming in faster than new reports. bpf, or Berkeley Packet Filter, is a Linux kernel technology. Torvalds called the development “lovely if true,” but cautioned that most areas have not reached that point. He said it could be a sign that “we may be starting to see the tide starting to turn,” while warning readers not to “hold your breath.”
Kroah-Hartman is closer but less certain. “We are getting closer, yes, but not quite there yet,” he said in an interview, and he would not estimate when human-approved AI fixes will overcome AI-generated bug reports.
Daniel Steinberg, creator of the open-source cURL project, has been critical of AI slop bug reports but is also cautiously optimistic. He said in an interview that cURL’s maintainers “manage the flood pretty well so far, even if it goes a little up and down in periods depending exactly on who sends us a new set of findings this particular week.”
What readers should watch next
For users, the short-term takeaway is practical: more AI-assisted vulnerability reports may make patch cycles look busier without making every report equally urgent. The bottleneck is human review, and that can delay fixes, especially in projects with few maintainers.
The source also warns that serious security problems are likely to continue. In November alone, it cited a serious Linux TCP networking security hole, an Alpine Linux OpenSSH flaw, Alpine’s GitLab near-miss, and the wide-open PHP2Shell security problem, which many groups still had not patched.
The balance may improve if AI-generated fixes keep accelerating and if projects gain more reviewers. Until then, the story is not simply that software has more bugs. It is that the cost of finding bugs has fallen, while the cost of trust — the human work needed to review, patch, and ship a fix — has not.

Comments (0)