Linus Torvalds has a message for anyone who doesn’t want AI in Linux

If you feel uneasy because AI has crept into the software you use, Linus Torvalds has a message for you. This is the man behind Linux, and he’s never been known for sugarcoating. I found his answer extremely eye-opening. It goes far beyond kernel code and is still relevant even if you’ve never touched Linux.

Fork it or walk away

Torvalds’ reply to Linux’s AI critics, and the rule behind it

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.

These were the exact words of Linus Torvalds on July 14 as he replied to a mailing list thread about anti-AI sentiment. His message is simple but to the point. If you don’t like it, copy the kernel’s code and run your own version, or simply “walk away.” This is quite blunt coming from a man with the final say on what ships in the official, mainline Linux kernel.

I followed the story and noticed that most of the reporting didn’t get past this point, but there were other important bits. He made it clear that no one can be forced to use AI, but he’d loudly ignores anyone who attempts to dissuade others from using it. In other words, he is targeting anyone who feels they should police someone else’s tools.

Just three months after this statement, he backed up his stance when a kernel developer wanted contributors to include an Assisted-by tag as a means of declaring AI use. Torvalds’ response was to tell him to stop using compilers for his submissions, and also skip his text editor since backspace hides typos. His argument was simple: people should understand what they submit, and the tools they use have little to do with it.

Once I put these two replies together, the rule became clear: The test isn’t the tool but whether you can stand behind the work.

From “ignore it” to “fork it”

Linux’s most famous AI skeptic flipped in under two years

It’s worth bringing some history to light. In just about 21 months, it looks like Torvalds’ stance has evolved. In October 2024, in an interview with TFiR, he said, “My approach to AI right now is I will basically ignore it.”

Something has genuinely changed. Greg Kroah-Hartman, who is a veteran kernel maintainer, said just in March 2026 that AI bug reports had gone from low-quality to genuinely useful. Then, in August 2026, he talked about how finding bugs is a job that is suited to AI. His reasoning was that AI is a pattern matcher and code is nothing but patterns.

By then, Torvalds was referring to the unusually large Linux kernel test releases, the “new normal,” and many of the extra fixes coming from AI-assisted code review. He also suggested that only people who have not tried AI would doubt its usefulness.

There is also a human detail to it. Torvalds built a guitar pedal in C, but needed to redesign the interface so that it looked modern. He said, “I don’t do Java, I do kernels,” and added he uses AI for the things he’s bad at. Like many of us, the Linux creator reaches for AI to help with things outside his skill set.

Not an AI free-for-all

Humans still decide, and AI is already straining them

Before you picture Torvalds as a guy who now processes everything with AI, consider his experience from March 2026 recorded on The Register. A release candidate came in too big, and his observation was that AI code review had resulted in several patch series. When the focus needed to be on regressions and stability, minor, non-urgent fixes were arriving late in the release cycle.

He said he’d become a bit more hard-nosed about that kind of churn. Just a week prior to this, he complained that the security mailing list had become almost unmanageable because of AI-found bugs; several people running the same tools were reporting the same problems.

Then, just in October 2026 on stage in Prague, he admitted that AI is stressing maintainers out to the point that it’s a problem. The day before, close to three quarters of the talks at the Maintainers Summit centered around making AI tools less stressful to deal with.

Through all this, I see one imbalance. Patches and bug reports are now cheap to produce because of AI. However, human time is still spent on checking and deciding which suggestions are correct, useful, and worth acting on.

Each project’s AI policy helps determine three important elements: who carries that extra workload, what gets filtered out, and what maintainers are expected to deal with.

The AI-free exits

There are several projects that have gone the opposite direction, so there is a place to go if Torvalds’ remarks leave you cold.

COSMIC’s reasoning is that AI-generated contributions need careful review and mostly come from people who don’t know the project.

I still take the licensing argument most seriously. Using AI doesn’t mean the patch can bypass GPL licensing requirements, and the contributor is still responsible for ensuring the submission complies. According to OpenBSD’s Theo de Raadt, understanding your code doesn’t mean you can trace what a model copied from somewhere else, and it’s impossible to guarantee that the code is yours to license.

Or you might simply apply Torvalds’ test: can you explain it to someone whose job is to say no? If you can, you’re on safer ground. If you can’t, there’s a problem.