Torvalds: oversized Linux kernel updates are the new normal, thanks to AI

Full text

Linus Torvalds called Linux 7.2-rc7 part of a "new normal" of oversized release candidates

Torvalds attributes the volume of fixes to reviews by various AI tools, a week after calling rc6 the biggest in years by commit count

The growth is in late-cycle bug fixes rather than features, and Torvalds credits AI review rather than AI authorship, with humans still writing and submitting the patches

Modern coding approaches dictate that Release Candidates should be expected to shrink as patches and code fixes get increasingly smaller as the software enters what many call a "stabilization phase".

Linux 7.2-rc7 did not do that, and neither did rc6 before it - and announcing the seventh candidate on August 9, 2026 Linus Torvalds wrote he could not say he was thrilled about the size of it, but that it is what it is: the new normal, with a lot of fixes, many of them the result of review by various AI tools.

A week earlier, opening the rc6 message , Torvalds had been blunter still, calling that release candidate huge and reckoning it the biggest rc6 in years by commit count.

An easier, detailed AI-enabled review process?

Torvalds is not describing AI writing for the Linux kernel; that still requires human intervention. The tools being used are review and analysis agents pointed at existing code, and they produce actionable bug reports. Humans then write the fixes, submit them through the normal subsystem maintainer process, and the creator of Linux, as the maintainer, pulls the code.

Phoronix, which closely tracks the cycle, describes the current environment as one in which AI and LLM coding and review agents keep kernel activity at an all-time high. It has resulted in a much larger-than-normal Release Candidate this late in the cycle; RC7 comes packed with a plethora of bug fixes that would otherwise have been in play much earlier in past cycles.

By one count of the RC7 pull, more than 400 fixes came in from upwards of 230 contributors, a volume more typical of an early merge window than a closing candidate.

Torvalds noted in the accompanying message that nothing looked particularly scary, and that the majority of the fixes were spread across drivers, filesystems, core networking, and architecture code, which was relatively small, making a potential release by the following weekend possible.

The AI approach doesn't have Mr. Torvalds particularly keen , but he has subtly acknowledged its advantages when it comes to code review, and it could bring a different issue into play: the maintainer model, which involves a few trusted individuals to read and test code before it is eventually released to the masses may not be too keen to have the equivalent of an AI-powered firehose of fixes and new information to deal with indefinitely.

For now, it seems larger updates are the new norm, for better or worse, and AI will be powering an increasingly large share of code review in an ecosystem that drives most of the world's servers, smartphones and tablets, as well as a sizeable portion of its personal computers.