+-----------------------------------------------+
|The views and thoughts below are purely my own.|
+-----------------------------------------------+

This post was written 2026-03-01

Claude Is Not Your Co-Author

…and related thoughts about LLM usage in open source.

Everyone has a take. You have a take; your friends have takes; the guy who wants to strike up conversation at the grocery checkout has a take. Some are good takes. Some are bad takes.

This one is mine.

There’s a lot happening in open source communities as we grapple with the deluge of spam produced by text extrusion engines. In my habitual corner of the broader community, the Python ecosystem, more and more projects are being forced to adopt explicit policies.

Having a written policy for acceptable conduct is good. Being forced to write one under duress is not good.

One particular detail of these policies has jumped out at me, which is how we treat LLMs as they appear in commit Co-Authored-By lines.

If you’re not familiar with it, the Co-Authored-By annotation is a line in a commit message used to cite your co-authors. I use it all the time – I even have editor macros for adding co-authors. It’s a great way to share credit with your co-conspirators colleagues and peers. For example, here’s what a commit1 with a co-author looks like:

commit 819ea3ca6836026bc611d0c8ea1cd5e95cbefc09
Author: Savannah Ostrowski <savannah@python.org>
Date:   Sun Feb 22 10:43:35 2026 -0800

    Refactor jit.yml (#144577)
    
    Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>

Recently, it’s become common to see a non-human co-author cited in these, e.g.

commit blahdeblahhashgoeshere
Author: Stephen Rosen <sirosen0@gmail.com>
Date:   Friday, Feb 31st 11:00:01 2026 -0500

      Fix the thing (#404)

      Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Which would indicate that I wrote the commit using – and it’s not clear to what degree relying upon – Anthropic’s Claude. I’m not sure who came up with this, but I think it’s very clever (not necessarily in a good way) and a bit weird.

But first, let’s look at the policy angle, which can help us understand the landscape.

The attrs Python library recently added an AI policy document. In it, they ask contributors to explicitly remove any LLM co-author tags. The rationale for this is clear in the context of the policy: you are asked to take full ownership of the code you submit, and so you must do some explicit work to acknowledge that. It also means that attrs won’t have those co-author tags in its history, which may have interesting knock-on effects.2

At the same time, pytest is working on their own AI policy document. In it, they ask contributors to credit LLM tools via co-author tags. The rationale for this is sensible in the context of the policy: you may not have written each line of code yourself, and you should communicate that clearly and honestly with the maintainers.

And many policies don’t say anything at all about this. For example, the pip-tools one I just wrote, with a lot of help and community input, does not mention the co-authored lines, and neither does the SciPy policy which I used as one of my references.

In terms of my beliefs and values, I’m probably closer to the opinion expressed in the attrs policy than not. You shouldn’t have those co-authored lines: you wrote and submitted the code, regardless of what tools you used.

Do I mark every commit I make from vim with Co-Authored-By: Bram Moolenaar or Co-Authored-By: vim? Have you ever seen an artist sign their name on a painting? Did they sign it “co-created by Blick”?3 Do book acknowledgements typically feature a shout-out to Logitech for the author’s mouse? If you cook a dish for guests, do you bring it to the table with a “I hope you all enjoy this, I cooked with Cuisinart”? Think about it a bit. Isn’t it slightly wrong to call your tool a co-author on your work?

There are two things about the co-authored lines that I do not like.

  1. Just as co-authorship can be used to share credit, it can be used to subtly disperse responsibility.

  2. I consider those co-authored tags to be a supremely clever bit of guerrilla marketing. I don’t like having advertisements in my community spaces.

By ostentatiously showing off where their product is being used, Anthropic4 is able to further normalize its use and claim an undefined amount of credit for other people’s work. Not that there’s anything deeply wrong with marketing your product. I just don’t think the code artifacts of the open source community are the appropriate forum for that.

It’s possible that the pip-tools policy will adjust to have an explicit stance regarding these tags. It may or may not reflect my personal opinion. For now, I won’t demand that anyone changes what they’re doing.

But please, consider removing those tags. And if you don’t feel comfortable owning the code you’re submitting, don’t submit it.

Addendum

It has rightly been pointed out to me that pre-commit and dependabot, among others, have long used these same tags. To date, neither I nor anyone else that I know of has complained about that usage.

I find the nature of the tools relevant, in that nobody mistakes these automations for having any agency or ownership over the changes they produce. However, unlike pre-commit and dependabot, developers sometimes refer to LLMs as their collaborators – which clearly implies a different relationship with the tools.

If the outcome of my argument is that we must take an “all or nothing” approach, and excise these annotations for non-LLM tools from projects as well – arguing that Dependabot commits advertise GitHub, just as Claude commits advertise Anthropic – then I am comfortable making that tradeoff.

  1. Savannah is the release manager for Python 3.16 and 3.17, and Hugo is the manager for 3.14 and 3.15 . It makes sense that they’d collaborate, but really I just picked a random cpython commit from the past few weeks with two authors. ↩

  2. There has been some rather hostile interrogation of “which projects are using AI?” going about. Although it’s reasonable to be concerned about how projects are safeguarded against poor quality or stolen work, I find much of the public discourse to be making much too much of a weak signal. At any rate, attrs may be inoculated against some of this accusatory behavior. ↩

  3. My spouse rightly pointed out that “influencers” would do exactly this, as part of their sponsorship deals. I found this to be a ringing endorsement of my opinions. ↩

  4. It really seems like Claude is the only LLM which appears in commits like this, presently. I’m sure other vendors will start doing it too. ↩