Maintainers are moving to remove “Assisted-by” because they consider it an inconsistently used disclosure mechanism that no longer serves its intended review purpose—not because they are banning AI-assisted kernel development. The change leaves human responsibility for reviewing and certifying contributions intact.
7
8
1
Why the tag is being dropped
- Limited usefulness: The tag was intended to help maintainers review AI-assisted contributions, but only about 10% of the 18,533 non-merge Linux 7.3 commits through -rc7 carried it.
8 That measures declared assistance, not actual AI use; the available evidence does not establish the true percentage.
- A collective decision: A majority of roughly 30 invited developers supported removal at the October 8 Prague Maintainers Summit, and Sasha Levin posted the removal patch the following day.
13
5
6 The summit decision therefore preceded Torvalds’s October 13 public criticism; it was not simply a policy reversal triggered by that response.
5
15
- Human accountability remains: Official kernel documentation explicitly prohibits AI agents from adding
Signed-off-by lines, because only humans can certify the Developer Certificate of Origin, and makes the human submitter responsible for reviewing AI-generated code.
1
2
Timeline and the disclosure dispute
- Introduction: The earlier search results describe a December 23, 2025 proposal approved by Jonathan Corbet in January 2026, while the versioned Linux 7.0 documentation confirms that AI-assistance guidance existed for that version.
13
2 This explains why accounts may describe the policy as originating in December but being adopted in January.
- Implementation status: Reporting dated October 16 says the removal consensus had not yet been fully implemented.
13 Insufficient evidence in the supplied excerpts to independently confirm the more specific claim that Corbet had accepted it into docs-next but not mainline.
- Torvalds’s objection: Reporting quotes his October 13 response as, “Do you not see how UTTERLY STUPID those kinds of rules would be?”
15 However, the supplied excerpts do not establish the full context of Laurent Pinchart’s objection or Torvalds’s compiler analogy, so that quotation should not be presented as rejection of every form of AI disclosure.
- Remaining disclosure: Removing a commit-message trailer is not logically equivalent to removing disclosure from submission discussions. Insufficient evidence in these excerpts to verify the continuing cover-letter requirement or Jeff Layton’s specific criticism that the tag advertises proprietary LLM companies.
What Torvalds’s own AI use illustrates
- Torvalds published an unmerged, eight-patch page-cache series using Claude Code; reporting describes roughly threefold improvement for small parallel reads on his laptop.
4
9
16 The cover letter reportedly credited Claude Code with code, test programs, and assistance with the write-up.
16
- Insufficient evidence here to verify the separate 23-fold synthetic-benchmark result, his real-workload caveat, or his October 9 remarks to Dirk Hohndel. The reported threefold result should not be substituted for, or conflated with, that claimed 23-fold measurement.
The practical implication is a shift away from a standardized tool-attribution tag toward evaluating the contribution and holding its human submitter accountable. That does not make AI-generated code automatically acceptable: the official review and human-certification obligations remain.
7
1