Linus Torvalds released Linux 7.1-rc5 on May 24. The candidate arrived bigger than past rc5 releases. He voiced clear displeasure with the volume of changes this late in the cycle.
Torvalds laid out his concerns in the announcement posted to the kernel mailing list. “To the surprise of absolutely nobody by now, rc5 is pretty big,” he wrote. “Quite a bit bigger than rc5’s have traditionally been.” The message continued with a blunt assessment. Most of the updates touched random drivers with trivial fixes. While low risk, they still carried some chance of disruption. He questioned whether such work belonged in late-stage release candidates at all.
The kernel chief signaled a shift in approach. He plans to grow more selective about pull requests from now on. Non-critical fixes to longstanding issues should wait for the next merge window, he argued. They do not qualify as regressions. Nor do they rise to the level of serious security concerns that demand immediate attention. “We are supposed to look for *regressions*,” Torvalds stated. This rc5, he concluded, had grown too large.
Phoronix first reported the release and Torvalds’ reaction. The site noted that several patch series stemmed from AI code review. Tools such as GitHub Copilot and Claude Code played a visible role. Their contributions touched graphics drivers, WiFi components, security issues in C code, and more. One follow-up article detailed specific areas helped by these agents. They included the Intel Xe driver, Raspberry Pi VD driver, AMD display code, SMB, Netfilter, sysfs, IO_uring, and Bluetooth.
Developers tagged many of these patches with “Assisted-by” credits. The practice highlights a growing presence of large language models in kernel work. Torvalds has acknowledged that AI tools can prove useful. Yet he drew a firm boundary this week. “AI tools are great when not causing unnecessary pain” and “pointless make-believe work,” he remarked in comments covered by Phoronix. The combination of AI-assisted patches and human-submitted driver tweaks produced a week of heavy activity.
Networking received another substantial batch of updates. One merge request carried a telling note. The craziness, it said, continues with no end in sight. Sound subsystem changes arrived in volume as well. Platform driver improvements expanded support for additional HP and ASUS laptops. Power management fixes addressed issues in both Intel P-State and AMD P-State drivers, including dynamic EPP handling.
But the size of the pull requests told a broader story. Linux 7.1 sits near 40 million lines of code. Earlier in the cycle, maintainers removed outdated ISDN and ham radio drivers, obsolete PCMCIA and PCI code, and began phasing out i486 support. Those deletions failed to offset new additions. The kernel crossed 30 million lines of actual code even as comments and blank lines pushed the total closer to 40 million. AMD’s graphics driver stack alone grew by more than 100,000 lines.
The rc5 episode fits a pattern visible across recent cycles. Merge windows pull in thousands of changes. Linux 7.1-rc1 closed the 7.1 merge window with roughly 13,000 non-merge commits. New features landed then. They ranged from a rewritten NTFS driver to expanded hardware support for upcoming Intel and AMD platforms. Real-time PREEMPT_RT support reached ARM architecture without external patches. Apple Silicon gained better battery reporting through the SMC power driver.
Stabilization phases normally grow quieter. Not this time. Torvalds’ frustration echoes earlier comments. He has warned that late churn undermines confidence in the final release. Small patches may seem harmless. Their cumulative effect raises the odds of unexpected behavior. And stability matters most for distributors and enterprise users who ship this code to millions of systems.
Industry observers took note. LWN.net covered the prepatch release within hours. It quoted Torvalds directly and highlighted the tension between fixing old bugs and protecting the release schedule. Linux-compatible.org described the rc5 as carrying an unusually large number of minor driver tweaks. The site reported Torvalds’ intention to reject noncritical requests going forward. The goal remains clear. Protect long-term stability over the urge to squeeze in one more improvement.
AI involvement adds a fresh dimension. Kernel developers have experimented with large language models for years. Some use them to spot potential bugs or generate boilerplate. Others apply them more aggressively. The “Assisted-by” tags in 7.1-rc5 mark a visible uptick. Whether this trend accelerates or meets resistance will shape future cycles. Torvalds’ message sends an early signal. Automation must not become an excuse for bypassing review standards.
So the clock ticks toward mid-June. Linux 7.1 stable sits weeks away. Testers continue to run the release candidates on servers, desktops, and embedded devices. Distributors watch for regressions that could affect their users. Hardware vendors verify driver behavior on new platforms. The community debates how strictly to enforce the rc-phase rules Torvalds outlined.
One fact stands out. The Linux kernel remains a massive collaborative project. Thousands of developers contribute each cycle. Their work spans filesystems, networking, architecture ports, and device drivers. AI tools now sit among those contributors. They accelerate certain tasks. They do not replace judgment about what belongs in the mainline at what time.
Torvalds has issued similar warnings before. This one carries extra weight because it ties directly to AI-generated or AI-inspired patches. The kernel’s growth shows no sign of slowing. Its line count keeps climbing despite periodic cleanups. That expansion brings new capabilities. It also demands tighter discipline during stabilization.
Developers who submitted patches for rc5 now face a clearer expectation. Ask first whether the change fixes a regression introduced since rc1. Or whether it addresses a critical security matter. If the answer is no, the patch should wait. The development tree exists for a reason. Late-cycle churn, even when trivial, risks introducing the very problems the rc process aims to eliminate.
The next few release candidates will test whether this message takes hold. If rc6 and rc7 arrive smaller and more focused, Torvalds’ stance will have landed. If the pattern repeats, further adjustments may follow. Either way, the conversation about AI in kernel development has moved from abstract possibility to concrete policy discussion.
Linux 7.1 already carries notable advances. Better file system options, improved power management, expanded hardware support, and real-time capabilities on new architectures all sit inside it. Those features will reach users soon enough. First the project must navigate these final weeks of testing without unnecessary distraction. Torvalds made plain his determination to keep that path as clean as possible.
WebProNews is an iEntry Publication