Linus Torvalds released the seventh release candidate of Linux 7.2 on 9 August and used the announcement to explain its bulk: "I can't say that I'm exactly thrilled about the size of this all, but it is what it is: the new normal with a lot of fixes, many of them due to review by various AI tools."
Why rc7 is the wrong place for this
An rc7 carrying roughly 400 non-merge commits across about 412 files is out of character for the kernel's schedule. By the seventh candidate the cycle has normally gone quiet, with maintainers holding anything that is not a regression. Torvalds made the same complaint a week earlier at rc6, which was the largest rc6 in kernel history.
What is in the pile
Not cosmetics. The batch includes a use-after-free race in memory management that had sat latent for eight years, HWMON fixes rated critical and high severity, a Btrfs fixup-worker restoration addressing silent data loss, a substantial block of networking patches and a fix for the Safe RET Interrupt vulnerability.
Review, not authorship
The claim inverts easily and several write-ups have inverted it. Torvalds is crediting AI tools with finding bugs during review, not with writing the kernel. He is also not saying quality fell — these are real defects, several of them years old, that human review did not catch.
An impression, not a measurement
"Many of them" is Torvalds' read of his own inbox. No commit-level accounting of AI-surfaced versus human-found fixes has been published, and the kernel's separate debate over attribution trailers has not settled. Final 7.2 is still expected next weekend.