Release candidates exist so a kernel can settle down before it ships. The final stretch of the Linux 7.2 cycle has done the opposite: the seventh candidate, posted on August 9, carried more than 400 fixes signed off by more than 230 developers, a volume that looks more like the opening of a development cycle than the closing of one. Linux creator Linus Torvalds is not pleased about the size, but he is treating it as the project’s new normal, and he will not let it delay the release, which is expected around August 16.
The reason for the growth is not a sudden surge in human-written code. Torvalds attributes much of the volume to AI tools reviewing the tree: automated systems combing through kernel code, flagging suspicious patterns, and generating bug reports that maintainers then triage. The distinction matters, because the flood has been widely misread as machines writing the kernel. What changed is that the machines are now doing a significant share of the looking, while humans still do the deciding.
A concrete case walked through rc7. A use-after-free flaw in ptdump, the utility that prints the kernel’s page tables, had sat undiscovered since March 2018, when it entered the tree with Linux 4.16. Syzbot, the automated fuzzing system that hammers the kernel with malformed inputs, flushed it out in June. A developer wrote a first fix with help from a large language model, the memory management maintainer reworked it, and the result landed in rc7 with plans to backport it to stable kernels, exactly the kind of latent bug that tends to become a vulnerability.
The same automation is pushing maintainers to delete code rather than let machines pick at it. Greg Kroah-Hartman removed the Moxa Intellio driver, roughly 2,200 lines written in 1999 for multi-port serial cards, arguing that no known hardware using it remains in circulation and that language models were starting to probe it for weaknesses that would waste everyone’s time. Obsolete code is being culled preemptively, before automated review turns it into a liability.
Torvalds’ relationship with AI in the kernel has been openly conflicted and consistently pragmatic. In January he objected to documentation being used as cover for poor AI-generated contributions. In May he complained that the kernel security mailing list had become almost unmanageable because of AI-generated bug reports, many of them duplicates. In July he declared that Linux is not an anti-AI project and told opponents of machine contributions to leave. The throughline: AI-assisted review is welcome, AI-generated noise is not.
The scale of the change shows in the numbers. The sixth release candidate, on August 2, was the largest rc6 in years by commit count, roughly 60 percent drivers, 20 percent networking, and the remainder split across architecture, tooling, and filesystems. RC7’s three larger fix clusters were s390 and zcrypt changes, the return of the btrfs fixup worker infrastructure, and netfilter ipset fixes. None of it alarmed Torvalds individually; the aggregate is what unnerved him, against a codebase that now exceeds 43 million lines.
The open risk is not today’s volume but tomorrow’s. The kernel remains manageable while its maintainers can still read what the tools surface. If the flood of machine-generated reports ever outpaces the humans who sort them, the project loses the ability to tell a real bug from a hallucination, and the quiet phase of the release cycle disappears for good.
Sources: Linus Torvalds says AI has made ‘huge’ Linux kernel updates the new normal (The Register, Aug 10, 2026); Linux 7.2 – AI reviews kernel code and Torvalds finds the bill steep (Korben, Aug 10, 2026); Linux Kernel 7.2 RC6 Arrives as Largest RC6 in Kernel History (Linux Compatible, Aug 3, 2026); Linus Torvalds Reaffirms That Linux Is Not “Anti-AI” (Phoronix)