Linus Torvalds released Linux 7.4-rc1 on Sunday, November 1, closing a two-week merge window he called "*BIG*": "almost 20k commits vs the more common ~15k", and "full of fixes for issues found by AI". He said he would rather see that flood of AI-found bugs fixed in the merge window, when new code is accepted, than in the weeks of testing that follow, and called it "a much healthier model".
A SaasBuzz count of the kernel's git history puts the window at 19,780 commits, or 18,591 excluding merges, about as many as the whole 7.3 release took in nine weeks. Torvalds also wrote that people expect 7.4 to be the next long-term support (LTS) release, a version kernel.org keeps patching with bug fixes for years.
What Torvalds said
Torvalds, who maintains the mainline kernel, wrote that the size had an "obvious" cause and that he took it as a good sign.
I'm hoping that this merge window being larger than usual is a sign that the AI fixes are becoming more part of the "normal development" than a flood of fixes that go into the whole stabilization phase.Linus Torvalds, Linux 7.4-rc1 announcement
He added that "more and more the patches have been suggested by AI too, although they often get rewritten", and that the expected LTS status is "another independent reason" some releases run larger: "So it may not *just* be the flood of AI patches." A third of the diff, he wrote, is "just AMD GPU register offsets and bit masks". Set those aside and the release is about half drivers, with the rest spread across architectures, filesystems, tooling and the core kernel. "But we'll see how the -rc's look," he wrote.
The numbers in context
Merge windows have grown every cycle this year, and the share of commits that carried the kernel's voluntary "Assisted-by" tag for AI help grew faster. SaasBuzz counted both in the kernel's git history for each window, from the previous release to its rc1.
Window | Commits | AI-tagged | Share |
|---|---|---|---|
7.1 | 12,996 | 134 | 1.0% |
7.2 | 13,412 | 706 | 5.3% |
7.3 | 15,267 | 1,257 | 8.2% |
7.4 | 18,591 | 3,008 | 16.2% |
Source: SaasBuzz count of non-merge commits in the mainline git tree, November 1, 2026. AI-tagged: commits with an Assisted-by line.
The 7.4 window is 42.6% larger than the average of the five before it, counting non-merge commits. Phoronix called it "the busiest on record" on a commit basis. The tag counts are a floor: Sasha Levin, whose change dropped the tag in this same window, wrote that "actual LLM use is far more widespread" than the roughly 10% of 7.3 commits that carried it.
The kernel's written rules for AI help
The kernel documents its expectations in two pages. The AI coding assistants guide says "AI agents MUST NOT add Signed-off-by tags", because only a person can certify the Developer Certificate of Origin, and the human submitter takes "full responsibility for the contribution".
The same page sets a procedure for bugs found with AI. The assistant must try to build a reproducer for any non-trivial bug, write and test a fix ("This part is not optional"), and say plainly what could not be done, because "maintainers currently waste too much time analyzing unverified reports and untested fixes." The assistant "must never send anything itself."
The tool-generated content guidelines ask contributors to say which tools they used and how, and warn that maintainers may reject a series "without detailed review" if the submitter cannot defend it.
The Assisted-by tag itself is gone. A documentation change by Sasha Levin, merged for 7.4-rc1, removed it after the 2026 Maintainers Summit, where "there was a strong, though not unanimous, consensus in the room to drop it". Disclosure moves to the patch series' cover letter, and the rules on sign-off and responsibility stay.
What it means for teams shipping on Linux
For companies that run their products on Linux, the AI-found fixes are concrete. The overlayfs pull Torvalds merged on October 31 lists "Fixes for bugs found by Chris Mason's AI review, several of them reachable by unprivileged users: an info leak, use-after-free and NULL deref bugs". OverlayFS is the kernel filesystem behind Docker's overlay2 storage driver.
Torvalds' argument is about timing. Fixes that arrive in the two-week merge window get the full seven weeks of release candidates to be tested before a final release. Fixes that arrive during those weeks get less. If the rc releases now stay small, the kernel will have absorbed a 22% jump in commits without a rougher run-in, which is what teams that track new kernels will be watching.
The LTS question raises the stakes. Torvalds wrote only that people "are expecting" 7.4 to be the next one. Kernel.org lists 6.18, released November 30, 2025, as the newest long-term kernel, with a projected end of life in December 2028. Whichever kernel is named next will carry this window's code, AI-found fixes included, for years.
For anyone sending AI-found bug reports, to the kernel or any project modeled on it, the kernel's bar is now written down: a reproducer, a tested fix and an honest note of what was not checked. A report without them, the guide says, "may cause the report to be ignored".
What's next
Release candidates come weekly. Torvalds said two pull requests he is "still thinking about" may land in rc2. Kernel.org describes a seven-week stabilization period in which rc7 is usually the last candidate, and Linux 7.0, 7.1, 7.2 and 7.3 each shipped exactly 49 days after their rc1. If 7.4 keeps that pace, the final release would come on Sunday, December 20. The Register also reported the announcement.
