Kernel maintainers move to drop Assisted-by AI attribution tag

Linux kernel developers are preparing to drop the Assisted-by tag that briefly required contributors to disclose when large language models helped produce a patch.

Sasha Levin posted a documentation change after the Maintainers Summit reached consensus to deprecate the tag. The tag had asked submitters to record LLM use, optionally naming specialized analysis tools, on the theory that attribution would track AI's role in kernel work. Contributors still had to add their own Signed-off-by line and take full responsibility under the Developer's Certificate of Origin.

Reviewers immediately asked for a fuller rationale in the commit message, noting the change reversed an earlier policy and that opacity would frustrate people on a sensitive topic. Levin withdrew the patch pending broader write-up of the summit discussion. Greg Kroah-Hartman backed removal but agreed the changelog needed to document the policy shift.

Support for dropping the tag has hardened among several maintainers. Kroah-Hartman said the experiment "didn't work out at all" and that maintainers are annoyed by it, arguing the project should focus on patch content and push back on anything wrong. Kees Cook called the tag as redundant as naming one's editor. Lorenzo Stoakes said the long-term answer to low-quality automated submissions is aggressive triage, not metadata.

Not everyone agrees. Steven Rostedt argued that git blame should still reveal when an author mostly prompted a model rather than writing the code, and floated a stronger marker for fully vibe-coded work. He noted he had not been at the summit to voice that view.

Jonathan Corbet is preparing a separate report on the summit conversation. The immediate effect of the change is simpler: kernel process docs and checkpatch will no longer require or police Assisted-by lines, leaving quality judgment to ordinary review.