While Torvalds Makes Peace With AI in Linux, Greg Kroah-Hartman Draws a Line (Sort of)
Greg Kroah-Hartman’s “clanker” branch started as his own experiment, where an AI-assisted fuzzing tool turned up real bugs in the ksmbd and SMB code. These days, he isn’t the only one using it, as other contributors have started sending in patches of their own, tagged the same way.
So when he posted a new AI policy a few days ago, restricting use of AI, it wasn’t him turning against a tool he’s comfortable with. It was a restriction that applied to one part of the kernel, the one place he doesn’t want AI anywhere near the code.
A new rule
In a post to the linux-staging mailing list, GKH said he’s now proactively rejecting LLM-generated patches for drivers/staging, pointing to an “onslaught” of them showing up lately.
His reasoning comes down to what staging is actually for. The subsystem exists primarily as a training ground, full of relatively simple cleanups and API changes that let newcomers learn the submission process.
The project already skips “fix all the coding style issues” patches generated by automated tools because clearing out that mess would eliminate the reason staging exists. GKH argues an LLM doing the same cleanup work defeats its purpose while calling out wannabe tricksters by saying that:
And yes, it is VERY obvious when people submit LLM-generated patches, so don’t think that just not disclosing the use of them will allow you to “get away” with anything here. The goal is for people to be able to learn, not to try to “trick” a maintainer. If anyone is determined to have deliberately “tricked” us, well, consider this your warning ahead of time…
There’s an exception, a strict one that too. A genuine security fix found by an LLM can still be submitted, but only if the submitter has tested it on the actual hardware the driver targets and can convincingly explain how testing was carried out.
A line drawn while Linus makes peace
This move lands right as Linus Torvalds is coming around on AI. He’s spent years shifting his stance, and in a recent mailing list post he made it clear to critics that Linux “is not one of those anti-AI projects,” telling them to fork the project or walk away if they disagreed.
GKH’s approach to drivers/staging fits inside the AI Coding Assistants policy, where he isn’t undoing what Linus signed off on but rather focusing on a narrower scope.
He is treating this subsystem as a gym, one where developers are supposed to show up and lift the weights themselves, not have an AI do the reps and hand them the result.
That’s the line GKH has drawn. AI can be used to carry out work elsewhere in the kernel, but in a subsystem built for people still learning, he’d rather they build it the hard way first.
💬 What do you think of GKH’s decision? Will it help newcomers?
Source: It's FOSS