Linux may drop AI-code labels — Torvalds uses LLMs, too

AI-assisted patches for the Linux kernel will no longer need an “Assisted by” tag in the commit message. A majority of the roughly 30 developers invited to the Maintainers Summit, held Oct. 8 in Prague, supported the change.

The consensus has not yet been fully implemented. But a corresponding patch has already been submitted to the Linux kernel mailing list. On Oct. 9, Sasha Levin, a distinguished software engineer at Nvidia and co-maintainer of the stable and LTS kernels, proposed removing the tag requirement from the documentation. Jonathan Corbet, the maintainer responsible for it, has since merged the patch into the docs-next development branch, but it still needs to make it into the mainline kernel.

The chances look good, because Linus Torvalds and Greg Kroah-Hartman, also one of the most prominent kernel developers, support the change. Opinions on the mailing list are divided, however.

Policy does little or nothing

Since Linux 7.0, the AI Coding Assistants policy has governed what developers must consider when submitting AI-assisted kernel patches. It is based on the consensus of the 2025 Maintainers Summit and a patch submitted Dec. 23, 2025, which Jonathan Corbet approved in January 2026.

Since then, developers have been asked to use an “Assisted by” tag in their commits when they use an advanced coding tool such as an LLM. Sasha Levin, who initiated the policy, has now argued that the tag should be eliminated. An initial proposal to remove it failed in July; Kroah-Hartman also supported keeping it at the time.

Levin explained why the tag should be eliminated in his revised patch from Oct. 13: Only about 10% of the 18,533 commits in Linux 7.3 through -rc7 carry the tag, excluding merge commits. In reality, AI is used much more often. According to Levin, some developers don’t know about the rule, others deliberately omit the tag, and hardly any maintainers ask about it. So a missing tag says nothing about how a patch was created. But for developers who use AI for almost everything, a tag doesn’t say much either.

The underlying goal of giving maintainers transparency into how a patch was created will remain. Developers will instead describe how they used AI tools in the cover letter, the email accompanying a patch or series of patches. That information no longer belongs in the commit message.

If the patch makes it into the mainline kernel, the Git history will no longer directly show which changes were made with AI assistance. But it was never reliable before, either. Responsibility remains unchanged: AI agents must not add a Signed-off-by line.

Linus Torvalds uses LLMs, too

Linus Torvalds has little regard for rules governing AI tools. He calls rules that ban the use of certain tools “stupid”. He expects people to understand what they’re doing, but says that understanding has little to do with the tools they use.

Around the time of the summit, Torvalds himself used an LLM to develop kernel code. In the commit message, he says the code, test programs, check used to compile the list of known issues, and first draft of the text were created with Claude Code under his direction. The patches are intended to significantly speed up reads from the page cache when many threads read small blocks from the same file simultaneously.

(mro)