The Linux kernel just absorbed one of its biggest batches of changes in a while, and AI is part of the story. On 1 November, Linus Torvalds released Linux 7.4-rc1 and wrote that it was a big merge window, with almost 20,000 commits compared with the more common 15,000 or so. In the same note he said the release is full of fixes for issues found by AI.
How a kernel release works
Every Linux release starts with a merge window, roughly two weeks in which maintainers send in the new code they have prepared. After that come weekly release candidates, which are mostly fixes, until the final version ships. The first release candidate, rc1, marks the end of the merge window, so its size tells you how much new work went in. Torvalds also noted that people expect 7.4 to be the next long-term support (LTS) release, the version that gets fixes for years, which tends to attract extra work.
What came in
Torvalds was specific about the role of AI in his announcement, and he still had two pull requests in his inbox that may land in rc2. He described the release as large and full of fixes for issues found by AI. He added that more and more patches have been suggested by AI too, although they often get rewritten before they are merged. That second detail matters, because it shows maintainers treating AI suggestions as a starting point that a person still checks and reshapes.
He was also careful not to blame everything on AI. He wrote that a full third of the patch, meaning the total size of the code changes, is AMD GPU register offsets and bit masks, which are large, mechanical header files. Because 7.4 is expected to be a long-term support release, he said the size may not just be the flood of AI patches.
Why Torvalds is pleased
The most useful part of the note is his reaction. Torvalds wrote that he is much happier to see the flood of AI-triggered work going into the merge window than to have it show up mainly as extra fixes during the release candidate period, and he called that a much healthier model. He also said he would wait to see how the release candidates look. The Register, which covered the announcement, reported the same points, although its headline says the kernel is "out of stabilization phase", a framing Torvalds did not use himself.
The timing logic is simple. Fixes that land early get weeks of testing before the release. Fixes that arrive late, during the release candidates, are riskier, because there is less time to catch the mistakes they bring with them.
What changed since the last release
This is a change from the previous cycles. In August, during Linux 7.2, Torvalds called the heavy volume of late fixes, many of them from AI review tools, "the new normal", according to The Next Web. When Linux 7.3-rc2 came out unusually large in September, he joked that people would blame AI anyway because it is an easy thing to blame. In both cases the AI-found fixes were arriving during the release candidates, which is exactly the pattern he now says he is glad to see moving earlier.
The kernel also has written rules for this. Its documentation on AI coding assistants says AI agents must not add the Signed-off-by tag, because only a human can legally certify where the code came from. The person who submits the patch must review all AI-generated code and takes full responsibility for it. That rule explains why AI-suggested patches "often get rewritten": a named human has to stand behind every line.
Why this matters beyond Linux
AI tools now help both find issues and suggest patches in the Linux kernel, a project where maintainers review every patch, at a volume big enough to change the size of a release. We think every software team can copy the kernel's approach of running AI bug hunts early and letting people rewrite the fixes. For an engineering lead, the lesson is about timing. For a security team, it is that AI bug reports are now arriving in volume, so triage capacity matters as much as the tools themselves.
What to do this week
Pick one module in a codebase you own. Ask your AI coding tool, whether that is Claude Code, Cursor or GitHub Copilot, to list likely bugs with evidence, then ask it to write a failing test for each one. Have a person rewrite any fix before it merges, record in the commit message that an AI tool was involved, and schedule this at the start of your release cycle so the fixes get tested properly. For help choosing a coding assistant, see our comparison page, and for a related story, read our piece on GitHub Copilot's model router.
Correction, 2 November 2026: An earlier version said a third of the changes in Linux 7.4-rc1 are AMD GPU register offsets and bit masks. Torvalds wrote that they make up a third of the patch, meaning the total size of the code changes, not a third of the commits.