The kernel's Assisted-by trailer makes AI help a signed line, not a rumour

Greg Kroah-Hartman pushed a set of new trailer keywords into the Linux kernel's sign-off conventions, including `Assisted-by` and `Inspired-by` ^[1]. The consequence is narrow and real: AI involvement in a patch becomes a line of structured text that a named human attaches, in the same commit-message field that kernel tooling already reads ^[4][7].

According to the dev.to write-up by Tamizuddin, the change landed in mid-2025 as an update to `Documentation/process/submitting-patches.rst` and to `scripts/get_maintainer.pl` ^[2]. The pre-existing vocabulary was `Reported-by`, `Tested-by`, `Reviewed-by`, `Suggested-by`, `Acked-by` and `Signed-off-by` ^[3]. One caution before you quote the list back to anyone: the same article names `Reviewed-by` and `Acked-by` variants among the new additions while also listing both as already available, so the exact boundary of what is new should be checked against the documentation rather than the summary ^[2].

The design decision worth copying is what `Assisted-by` does not do. It does not require disclosure of which tool was used, and it is placed by a person rather than emitted autonomously ^[4]. That puts it in the same category as `Reviewed-by` and `Acked-by`, which the article describes as social contracts rather than machine outputs ^[5]. Torvalds did not champion it; it came up through the maintainer chain as a practical acknowledgment that developers were already using AI and the existing trailers described the contribution badly ^[6].

This is cheap to adopt precisely because the plumbing exists. `checkpatch.pl`, `git log` and maintainer scripts already parse trailers for filtering and statistics ^[7], so a new keyword costs nothing on the parsing side and everything on the norm side. The article expects `Assisted-by` entries to eventually reach code intelligence dashboards, blame views and license or compliance audits ^[12], and expects corporate contribution guidelines to follow the kernel's tone ^[13]. Both are the author's forecast, not documented policy.

The line that will get argued about in review is the liability split. `Signed-off-by` certifies that you vouch for the code; `Assisted-by` records where help came from and does not transfer that responsibility ^[8][9]. The article's own advice is that if you accept AI output without review, only `Signed-off-by` is appropriate, and that you should reconsider the workflow ^[10]. That is the honest reading: the trailer is a provenance note, not a shared-blame clause.

Anyone planning to mandate this internally should note the gap. The form that names the model and the editor, such as `Assisted-by: Claude (Anthropic) via Cursor`, is not official kernel policy by the article's own admission ^[11]. Because the official tag requires no tool disclosure ^[4], a compliance process that reads only trailers cannot tell from the commit which model or vendor touched the code ^[1]. If that identification is what your auditors want, the kernel convention does not supply it and a local convention on top will not be portable.

What to watch: whether `checkpatch.pl` gains an enforcement or warning path for the new keywords, whether the tool-naming form is ever written into `submitting-patches.rst`, and whether maintainers outside the kernel copy the human-signed design or reach for something machine-generated instead. The second one decides whether provenance data is comparable across repos or just locally true.