AI tools now turn up software flaws faster than the people who maintain Linux and Windows can review and fix them, ZDNET reported.[1] In one recent case, MDASH, an AI scanning system from Microsoft's FORGE Lab, turned up 59 upstream Linux kernel bug reports, including one flaw at least 21 years old, according to ZDNET.[1]

Finding bugs is no longer the hard part, ZDNET wrote; reviewing and fixing them is, so the cost of cheap AI scanning falls on maintainers at Microsoft and in the kernel community.[1] Because Linux runs everything from smartphones to refrigerators to supercomputers, ZDNET noted, what may be a serious bug in one device may not even appear at all on another.[1]

ZDNET said Microsoft's FORGE Lab report that MDASH turned up 59 upstream Linux kernel bug reports is less impressive than it sounds.[1] For 2026, through Nov. 2, the Linux Common Vulnerabilities and Exposures (CVEs) team reported an average of 27 CVEs a day, and Microsoft reported 726 CVEs in its Patch Tuesday security release on Nov. 10, with the vast majority of such reports coming from AI programmers.[1]

The strain is showing, according to ZDNET: on Nov. 10, Lorenzo Stoakes, an ARM engineer and kernel contributor since 2014, gave up reviewing code outside the areas he maintains.[1] Stoakes wrote that AI-generated "slop" is now "pretty much half of what I see, if not more."[1] Greg Kroah-Hartman, the Linux kernel stable maintainer, said in a conference Q&A session, "We need more reviewers."[1]

According to notes from a May 24, 2014 kernel discussion published on kernel.org, James Bottomley pointed out that reviews were not keeping up with patches and Guenter Roeck objected that companies reward code production over code review.[2]

Torvalds wrote in a recent kernel mailing-list message that on the Berkeley Packet Filter side, fixes are now coming in faster than new reports, though he said most other areas have not reached that point, ZDNET reported.[1] Kroah-Hartman told ZDNET, "We are getting closer, yes, but not quite there yet," and would not estimate when human-approved AI fixes will overtake AI bug reports.[1]