The Linux kernel project has spent a considerable amount of time using AI tools, and the response has typically been “think for yourself” and “I will get back to you.. ”
At the 2025 Maintainers Summit late last year, Sasha Levin called for a written agreement. The result is that patches are non-negotiable, purely machine-generated submissions are not welcome, and the use of tools is a human responsibility for being publicly available.
He didn’t promise to enforce anything, but he did promise to document something, and the result is now shipped with Linux 7.0.
What is it?
Linux AI coding assistant policy.
The new documentation is called AI Coding Assistant and resides within the kernel process documentation along with the rest of the contribution guidelines. Simply put, AI-assisted contributions still only need to comply with GPL-2.0. AI agent cannot be added Signed-off-by Tags; Patches with the help of AI include:With assistance“tag.
Since the Developer Certificate of Origin (DCO) is an important element; Humans are responsible for all patches. AI assistance does not change those stringent requirements.
Essentially, the human submitter reviews everything the AI generates, makes sure it meets licensing requirements, properly notes that the AI was used, and submits his or her name.
of With assistance The format of the tag is Assisted-by: AGENT_NAME:MODEL_VERSION [TOOL1] [TOOL2];For scenarios where single or multiple tools are used. document give Assisted-by: Claude:claude-3-opus coccinelle sparse As an example.
At that time, Linus wasn’t even convinced that a dedicated tag was needed And suggested that the changelog body would do the job. But now the kernel community seems to have settled on this tag anyway.
already in use
We covered this earlier this week, and it looks like Greg Kroah-Hartman (GKH) has been running AI-assisted fuzzing in a branch of the kernel tree called “” for some time.PonkotsuStarting with the ksmbd and SMB code, he found several potential issues and submitted fixes with a note instructing reviewers to independently verify everything before relying on it.
This is about a workflow where a new policy is created.. AI surfaces issues, humans with decades of kernel experience determine what is true, write fixes, and take responsibility. It’s not surprising that GKH is doing that, considering they are maintainers of stable kernels and have probably dealt with more bad patches than anyone else.
Other projects went in different directions. Gentoo completely banned AI-generated posts in 2024, with its council citing copyright risks, code quality, and ethical concerns.
NetBSD’s commit guidelines require LLM-generated code to becontaminated code” category requires written approval from the core developer.
in contrast, Linux doesn’t prohibit anything. Whether it’s a wise decision or just a generous one depends on how seriously people actually take it. ”Human reviewed this” Some.
Recommended reading 📖: Is Clanker used in Linux development?
Is Clanker used to perform AI fuzzing in the Linux kernel?
Greg Kroah-Hartman appears to be running AI-assisted fuzzing on top of the kernel. This may not be a bad thing, so don’t get outraged yet.
