Artificial intelligence systems are increasingly used to identify software flaws, creating a bottleneck for developers who must review and fix the resulting flood of bug reports. This strain is particularly evident in both the Linux kernel and Windows operating system ecosystems, where maintainers report being stretched to their limits.
Microsoft’s FORGE Lab recently disclosed that its AI scanning system, MDASH, identified 59 upstream Linux kernel bug reports, including a vulnerability that had existed for at least 21 years. While this number appears modest, it contrasts with the Linux Common Vulnerabilities and Exposures team’s report of an average of 27 CVEs per day through November 2, 2026. The high volume stems from the fact that in Linux, any bug can potentially be exploited as a security flaw, as Linux creator Linus Torvalds has noted. Additionally, the latest Windows Patch Tuesday release included 726 CVEs, with the majority of recent reports attributed to AI-driven discovery.
The volume of AI-generated code and bug reports has put significant pressure on human reviewers. Lorenzo Stoakes, an ARM engineer and Linux kernel maintainer, ceased reviewing code in areas he did not directly maintain on November 10, citing a lack of bandwidth due to the high throughput of AI-generated series. Stoakes noted that AI-generated low-quality code has become increasingly common, stating that it now constitutes a significant portion of the code he encounters.
Greg Kroah-Hartman, the stable maintainer for the Linux kernel, echoed these concerns during a session at Open Source Summit Europe. He stated that the team lacks sufficient reviewers to handle the onslaught of patches generated by large language models, adding that many maintainers have been feeling burnt out for a decade.
Despite the challenges, some maintainers see signs that the situation may be improving. Linus Torvalds indicated that in some areas, such as the Berkeley Packet Filter, AI-driven fixes are arriving faster than new bug reports. Greg Kroah-Hartman remains cautiously optimistic, noting that the system is getting closer to managing the flood but is not yet there. Daniel Steinberg, creator of the cURL project, expressed belief that the community is managing the influx of reports effectively, despite periodic fluctuations.
Experts anticipate that serious security breaches will continue to emerge as the volume of unpatched AI-reported vulnerabilities remains high. Recent examples include a critical Linux TCP networking vulnerability, an Alpine Linux OpenSSH flaw, and the unpatched PHP2Shell security issue.
Seven hundred CVEs in one update? My patch management team is literally crying right now. This is insane.
Burnout after ten years sounds like a generational issue, not just an AI one. We need more paid reviewers.
I didn’t realize the AI-generated code itself was becoming a major problem for maintainers, not just the bug reports.
Finding a twenty-one year old bug is impressive, but the sheer volume of noise is drowning out the actual signal.