Linus Torvalds has had enough. The Linux kernel’s chief maintainer signaled this week that he will start rejecting trivial pull requests that arrive too late in the development cycle. Some of those come from developers spurred by AI code reviews. The move comes as the fifth release candidate for Linux 7.1 landed on May 24, 2026, and Torvalds called it “too big.”
“So I think I’ll start being a bit more hardnosed about this kind of unnecessary churn this late in the game,” he wrote in his weekly kernel update posted to the Linux Kernel Mailing List. He made clear the timing matters. These late-stage weeks focus on hunting regressions. Not on minor fixes to longstanding issues that could have arrived earlier.
The Register first reported the comments hours after they appeared. (The Register)
Torvalds didn’t mince words. Several of the patch series in question “were triggered by AI code review.” He sees the pattern. Tools scan old code. They flag potential problems. Enthusiastic contributors then rush fixes that don’t address current regressions. The result? Extra noise. Bigger release candidates. And risk that real problems get lost in the shuffle.
The Cost of Churn at the Wrong Time
Late-cycle volume creates headaches. Torvalds explained that large rc weeks do not support long-term stability. Even fixes with a low chance of causing trouble carry some risk. One bad interaction in a complex system can surface weeks later. Maintainers already spend hours triaging duplicates and redirecting reports.
This isn’t new frustration. Back in January 2026 Torvalds shut down debates over labeling AI-generated code in kernel documentation. He called such discussions “pointless posturing” because bad actors won’t follow the rules anyway. (Slashdot) His stance then treated AI as one tool among many. Humans remain responsible for what they submit.
By May the problem had shifted. AI tools now drive reviews that prompt patches at the worst moment. Neowin noted Torvalds’ growing impatience with non-critical changes bloating the cycle. He wants developers to check their work carefully. Does this fix a regression? Or does it belong in linux-next for the next merge window? (Neowin)
Phoronix tracked related developments throughout the 7.1 cycle. One report highlighted how AI coding agents contributed fixes this week while Torvalds warned that such tools prove great only when they avoid pointless work. The site also covered earlier removals of obsolete drivers partly because AI-generated bug reports flooded maintainers with noise on dead code. (Phoronix)
But Torvalds hasn’t rejected AI outright. In mid-May he merged new guidelines for handling security bug reports that mentioned AI tools. He acknowledged they can surface real issues. The catch? Contributors must do the engineering work. Read the docs. Understand context. Submit a proper patch instead of dumping raw scanner output. Recent coverage from IT-Connect detailed how the security list became “unmanageable” from redundant AI reports. (IT-Connect)
The pattern repeats. AI lowers the bar for finding potential bugs. It does not automatically raise the quality of contributions. Duplicate reports waste time. Trivial fixes at rc5 add risk without clear benefit. Torvalds wants accountability. The human who clicks send owns the code.
Maintainer Reality in an AI-Filled Inbox
Kernel development runs on trust and scrutiny. Torvalds has long rejected patches that fail basic tests or introduce new problems. His blunt feedback appears in public because transparency serves the project. That style now targets a modern source of noise.
Earlier this year the community settled on “Assisted-by” tags for AI help. The human signer stays fully accountable for bugs, security flaws, or licensing problems. No political statements in the docs. Just practical rules for good actors. Yet as Torvalds noted in January, the people generating low-quality “slop” won’t label it anyway.
So the burden falls on reviewers and ultimately on him. Bigger rcs mean more patches to read. More chances for oversight. More work chasing regressions that matter to users and enterprise deployments. Linux powers everything from smartphones to supercomputers. Stability counts.
Recent discussions on X echo the tension. Developers experiment with AI for code review. Some role-play Torvalds-style critique to catch issues early. Others worry the flood will force stricter gates. One thread from mid-May captured the shift: AI agents help with driver bugs but only if humans verify every change.
Torvalds’ latest warning serves as notice. Submit early. Target real problems. Use linux-next for cleanup. The late cycle stays reserved for what breaks today. Not what an AI flagged last night.
He built the kernel on a meritocracy of careful work. That hasn’t changed. AI can accelerate discovery. It cannot replace judgment about when and what to send. The head-up is clear. Pointless churn gets pushed back. Hardnosed review returns. And the project keeps moving.
WebProNews is an iEntry Publication