Repository navigation
AI contribution policy #9409
Description
Hi! Thanks for making borg, I've used it for a while (although not long enough for it to have a chance at saving my ass yet).
I wanted to know what's your viewpoint on using LLMs and accepting or refusing AI contributions. I'm a bit worried about what accepting (explicitly or not) AI generated code could entail, for many more reasons than just code quality. There's, as far as I'm aware, no clear guideline on this topic, so i think it could be a good thing to add this somewhere in the documentation (here?).
Thanks!
Activity
ThomasWaldmann commentedon Feb 24, 2026
I understand (some of) your concerns. I even share some concerns about using "AI" and I hate it if it is abused (see AI slop on YouTube that wastes everybody's time and also a lot of resources), but I am not generally against it. As often, "it depends". It is a tool, one can use it for good or for bad.
In general, contributions (from any source) are reviewed for good quality: code, style, tests (correctness), documentation, ...
Sidenote for people who are not software developers: "tests" here mean automated tests that run Borg code and verify if the code behaves as expected. If it doesn't, we would immediately notice, because these tests are run even before we accept a contribution (see GitHub Actions). We have such tests for all Borg commands and also for the inner parts of the code.
"No AI policy": There would not be much point in demanding something that could not be verified by looking at the contribution. Before AI, people used search engines, Stack Overflow/Reddit, or random websites and copy-pasted code of varying quality. One generally does not see where code comes from, but one can review the code and have tests for it, ensuring it works correctly.
Thus, contributions must be reviewed and judged based on their contents. It's not just AI that can write incorrect code, humans do that since ever, too. :-)
If AI can be used so that the few Borg contributors can be more productive (while maintaining good quality of course), it should be used. It's not like we have a thousand active Borg developers here waiting for stuff to do...
ANKITAPRAMODPATIL commentedon Feb 27, 2026
Hi, I am a beginner contributor and interested in working on improving the documentation regarding AI contribution policy.
Can I work on this issue?
ThomasWaldmann commentedon Feb 27, 2026
Hi @ANKITAPRAMODPATIL! Sure, just do a PR.
cyrneko commentedon Feb 27, 2026
AI-Generated code may not be copyrighted under US law, as per US Congress: https://www.congress.gov/crs_external_products/LSB/PDF/LSB10922/LSB10922.10.pdf
This also means AI-generated code may not be licensed.
Importantly, if someone fails to disclose what is AI-generated and what isn't, the entire Work may not be copyrightable.
ThomasWaldmann commentedon Feb 27, 2026
@cyrneko Sounds "interesting", especially considering that a lot of commercial software companies announced that they use AI. :-)
ThomasWaldmann commentedon Feb 27, 2026
From claude, answering "If AI is used to generate code, is the code copyrightable and licensable?":
This is a genuinely unsettled area of law, but here's where things stand:
The core issue: human authorship
Copyright law in the US (and most countries) requires human authorship. The Copyright Office and courts have consistently held that purely AI-generated works — with no meaningful human creative input — are not copyrightable. This was reinforced in cases like Thaler v. Perlmutter (2023), where a court rejected copyright for an image generated autonomously by AI.
But "AI-assisted" is different from "AI-generated"
In practice, most AI-generated code involves a human who:
The more human creative judgment is involved, the stronger the claim to copyright. The Copyright Office's current guidance is that it will register works with sufficient human authorship, even if AI tools were used — similar to how using Photoshop doesn't undermine your copyright in a photo.
Licensing
Yes, AI-generated or AI-assisted code can be licensed, though with caveats:
Practical takeaways
Courts and regulators are still working this out, and the law varies by jurisdiction. A few practical points worth knowing:
The short version: AI-assisted code where a human made meaningful creative choices is likely copyrightable and licensable. Purely AI-generated code with no human creative input is in murkier territory, and may not be protectable at all.
cyrneko commentedon Feb 27, 2026
That is, effectively speaking, a re-hash of what the 7-page document from Congress said.
To reiterate, if someone cannot disclose where AI was used -- just like your LLM rehashed -- then something may not be license-able.
Regardless, I don't like the take that AI contributions are only about licensing and code quality, there is so many issues with LLMs that go outside of that too that make me completely avoid them.
To be the one to break the ice here, the datacenters these models run on are often built in places with marginalized communities that are often also financially struggling. There, they raise the prices for utility/electricity bills as well as polluting the air, which means more people have to pay hefty fees for more visits at various health institutions as well as suffering in health.1
To look past these issues is, in my opinion, ignoring a larger issue that goes beyond code quality and speed of development. I would love to see Borg (and other projects) take a stance against these practices. People that continue to submit PRs and/or code with these tools are choosing to look past those moral issues in the name of a slight speed-up.
It does not have to be activism, but disagreeing with these practices and discouraging use of Generative AI (LLMs, image/video diffusion models) in PRs would be nice and, in my opinion, an important step.
Footnotes
https://conimby.org/blog/2025/09/12/the-real-safety-risks-of-data-centers-what-local-communities-need-to-know/ ↩
tasiaiso commentedon Apr 8, 2026
Hi, I'm wondering if there are any updates on this subject and if you'd be willing to accept a policy stating against the use of AI in Borg (personally if i had it my way i would outright ban AI contributions, but that's not my project nor am i a contributor or plan to become one in the near future, so i recognize i don't have much of a say in this decision)
contributing guide incl. AI policy, fixes borgbackup#9409
raingloom commentedon Aug 17, 2026
@tasiaiso @cyrneko I wanted to find an answer to this question too, so I did some digging.
The current top 5th contributor is Claude and most non-merge commits by @ThomasWaldmann are "co-authored" by Claude
The commit that closed this issue pushes the responsibility to mitigate the ecological and social harms onto the users.
Only @ThomasWaldmann knows if he actually adheres to those rules and whether the
Co-authored-byis warranted on every commit, even simple ones like this.As for other tools in the borg sphere (Cube? IDK, I never watched Star Trek.):
PIka Backup takes a firmer stance.
I haven't looked at Borgmatic's stance on LLMs, but they explicitly point out where you money goes if you choose Amazon for object storage, which in some ways is even better.
The Vorta project also allows and uses LLMs.
The new experimental Vykar tool by Borgbase is even more lax about LLMs. (And if I'm allowed to be a bit snarky, this was not surprising after reading their docs.)
I understand why these tools are tempting and I can empathize with people who feel like they have no other choice, especially when they don't know all the harms these "tools" cause — the surface of which we barely scratched.
But I do think the short term and small scale benefits are far outweighed by the larger scale ecological and social disaster that is unfolding.
At the minimum the deskilling of developers is a cause for concern. If the developers can't work on these projects without relying on one of a handful of megacorporations, then the projects are effectively dependent on those corporations. And it is clear that these models are getting more expensive and that hardware is unlikely to get cheaper (minerals are not getting more abundant), so even a purely selfish LLM user must ask themself what price they are willing to pay.
ps.: No LLMs were used to write or research this, I just know how to configure caps-lock as a compose key, hence the single em-dash. 😅
witten commentedon Aug 20, 2026
FWIW, borgmatic's generative AI policy is in its docs. It's probably closest to Pika's stance.
As for Borg, I think it's a little tricky. IMO any open source project maintainer should have the ability to decide how to govern and maintain their own project. And of course users should provide feedback on any such decisions—which is exactly what you're doing here. But if ultimately you disagree with a decision, and the decision is not reversed, then it's up to you whether you want to continue using the project.
Hund commentedon Oct 20, 2026
Well. This sucks.
Feel free to share links to any future non-AI forks of Borg here.