Linus Torvalds Blames AI for a ‘Full Fat’ Linux 7.3-rc2 – And Half the Community Isn’t Laughing

Linus Torvalds released Linux 7.3-rc2 on a recent Sunday. The volume surprised even him. This second release candidate usually offers a quiet breather after the merge window. Not this time.

“This didn’t *feel* like a particularly busy rc2, but it clearly was,” Torvalds wrote in his announcement. He called it a “full fat” release. Then came the punchline. “It might be just random. But we’ll obviously all blame it on AI, because whether that’s really the cause or not, it’s an easy thing to blame ;)”

The wink emoticon said it all. Torvalds has joked about AI before. This time the humor lands amid a serious shift in how the kernel gets built. Hundreds of patches arrived. Filesystems sent fixes. DRM contributed a sizable pull. Networking, BPF, and various drivers added their share. Nothing looked particularly odd, he noted. Yet the total reached 615 commits from 292 contributors, according to The Next Web.

But the joke carries weight. For months Torvalds has defended AI tools against critics inside the community. He has called them useful. He has told opponents to fork the project or walk away. Now an oversized rc2 becomes the latest exhibit in an ongoing debate. Is AI accelerating development? Or is it flooding the process with noise?

Torvalds has watched the numbers climb for over a year.

In May he warned that AI-driven bug reports had made the kernel’s private security list “almost entirely unmanageable.” Duplication ran rampant. Maintainers spent hours redirecting reports of issues already fixed. The documentation was updated to treat AI-detected bugs as public by default. No more private handling for what multiple bots would rediscover anyway. That post came in the 7.1-rc4 announcement, as covered by GamingOnLinux.

By July his tone sharpened. On a mailing list thread about an AI code review system named Sashiko, Torvalds stepped in. “Linux is not one of those anti-AI projects, and if somebody has issues with that, they can do the open-source thing and fork it. Or just walk away,” he wrote. The full statement appears in Ars Technica.

He continued. “AI is a tool, just like other tools we use. And it’s clearly a useful one. It may not have been that ‘clearly’ even just a year ago, but it’s no longer in question today.” The kernel focuses on technology, not ideology. Torvalds said he would ignore arguments against others using the tools. Phoronix published the complete exchange.

His own experience with the technology has been mixed. In August Torvalds credited AI with helping fix a stubborn Intel Xe driver bug. The session stretched across 24 patches and 18 kernel boots. The AI suggested several times that the problem might prove unsolvable. Torvalds pushed on. The final fix changed a round_up() to round_down(). He let the AI write the commit message. Details emerged in The Register.

So when rc2 ballooned, the AI reference felt natural. Yet the kernel project has already created mechanisms to track this influence. Contributors are asked to add an “Assisted-by: LLM” tag to commits that used machine assistance. The policy exists. Few seem to publish the resulting totals. That absence makes Torvalds’ joke land with extra irony, The Next Web observed.

Recent data points to real pressure. CVE counts per release have jumped. Releases from 6.9 through 6.19 averaged around 500 vulnerabilities fixed. From 7.0 onward the figure climbed toward 1,000, then past 1,500 with 7.2. Linux 7.3 could exceed 2,000. Most new findings come from AI code review platforms and large language models. Greg Kroah-Hartman highlighted the trend ahead of Kernel Recipes 2026. TechSpot reported the numbers.

And. The reports keep coming. Some point to already-patched code. Others target deprecated drivers with few users. Maintainers triage the flood. The volume creates churn. It also surfaces genuine issues that might have lingered.

Torvalds has described a love-hate relationship with the tools. He finds them useful. They expose embarrassing bugs in human-written code. They change how developers work. In a 2025 conversation with Dirk Hohndel at the Open Source Summit, he noted an uptick in development activity once AI tools improved. The discussion was summarized by LWN.net.

His stance remains consistent. No one must use AI. But opposition to others using it will be overruled. The project measures success by technical merit. A bigger rc2 doesn’t automatically signal trouble. It could reflect faster iteration. It could simply reflect random variation after a busy merge window.

Yet the pattern grows hard to ignore. Late EDAC pull. Filesystem fixes. Large DRM updates. Networking and BPF work. Driver trees across the board. Torvalds listed them all. He found no single culprit. So he reached for the explanation on everyone’s mind.

Phoronix noted the release also brought cache-aware scheduling fixes for hybrid CPUs, tree-wide kmalloc() cleanups to kmalloc_obj(), NVIDIA Blackwell display support in Nouveau, and a change to disable RandStruct security by default when Rust is enabled. Those changes arrived alongside the usual flow of driver and subsystem patches. The site covered the details in its report on the rc2 release.

The Linux kernel remains one of the largest coordinated software efforts on the planet. Nearly 41 million lines of code. Thousands of contributors. Releases every few months. AI has inserted itself into that machinery. Some see acceleration. Others see noise. Torvalds sees a tool that has crossed a threshold of usefulness.

His latest joke may prove prophetic. Or it may simply mark another Sunday when the patch volume refused to cooperate with the calendar. Either way, the conversation about AI’s place in kernel development has moved past theory. The evidence arrives weekly. Sometimes in the form of a bigger-than-expected rc2.

Developers will test this candidate. Bugs will surface. Fixes will arrive. And the question will linger. How much of the new activity traces back to machine assistance? The tagging policy exists to answer exactly that. So far, the community hasn’t tallied the results. Perhaps the next rc will bring that data. Or perhaps everyone will just keep blaming AI. It’s easy.