Add a statement on usage of AI tools

[eduardo-rodrigues]

It seems like a good idea to introduce a statement on the usage of AI tools based on recent experience from recent community contributions and on discussions held within the GSoC programme mailing list. Maybe a simple pull request template will do the job? Or a subsection on the README?

Comments and ideas welcome.

Ping @wdconinc, @maszyman, @vvolkl, @aaronj0 for input from our GSoC coordinators.

[wdconinc]

How about an organization-wide pull request template (https://github.com/HSF/.github), as .github/pull_request_template.md with a simple checkbox?

<!-- Describe your pull request at the top, and take note of the checkbox below -->



### Please check if any of the following apply
- [ ] AI was used in preparing this PR. Please describe usage below.

[vvolkl]

In principle, I like what the linux kernel now has: https://github.com/torvalds/linux/blob/master/Documentation/process/coding-assistants.rst

But that should probably go to CONTRIBUTING.md

[elmsheus]

In case it's useful, the following link was shared on the FOSS at CERN mattermost channel a few weeks ago: https://redmonk.com/kholterhoff/2026/02/26/generative-ai-policy-landscape-in-open-source/

[caterina-doglioni]

[note: I went to the HSF Github to tick off one of my TODO items and found that @eduardo-rodrigues had already done that (thanks!) so I'm spending a bit of time on this instead]

For our GSoC project we used this (with mixed success):

Use of generative AI: We would recommend you do not use generative AI (ChatGPT, copilot, etc) for this warm-up exercise, but if you do so, you must declare it and share your prompts using this declaration: GAIDeT Declaration Generator
If you overstate your own skills and abilities it will mean that we will not be able to work together towards a successful outcome of this project.

Which could become a bit more friendly - but also we should think how this changes things for our own use:

Use of generative AI: in the HSF, we enjoy interacting with our human contributors. We also understand that generative AI tools (e.g. ChatGPT, GitHub Copilot, Claude, etc...) are widely used - if you use one of these tools when contributing to our activities, please consider the following guidelines:

• take full responsibility for your pull request, and this includes understanding and carefully checking the output of the GenAI tool you are going to include;
• declare what tool was used; here we could add the PR box from @wdconinc, or another automatic procedure
• prior to generative AI use, consider whether:
  • your use of AI tools is clearly adding value to the HSF's activities;
  • the use of AI tools in your pull request shifts too much of the reviewing burden on the HSF;
  • whether its environmental costs [1] are proportionate to the benefits.

We will close pull requests that do not adhere to these guidelines.

[1] How Hungry is AI? Benchmarking Energy, Water, and Carbon Footprint of LLM Inference, N. Jegham et al.](https://arxiv.org/abs/2505.09598), HuggingFace primer on environmental impact of AI,

[eduardo-rodrigues]

Thank you for all this info, @caterina-doglioni 👍.

In the end many of us have been loaded recently, and that's the reason why I did not push for this task to be finalised. But it is in my TODOs for just after CHEP. I feel we have enough material above to finalise with the SG and AA conveners and anyone interested in contributing.

[kreczko]

For FAST-HEP packages, I have this blob (in CONTRIBUTING.md) - hope it is helpful:

---

AI-assisted contributions

AI-assisted contributions are welcome.

However, contributors remain responsible for the code they submit.

Contributors should:

• understand the generated code
• review generated changes carefully
• validate correctness locally
• ensure tests pass
• ensure the implementation matches FAST-HEP design principles

Pull requests containing substantial AI-generated or AI-assisted changes should include:

• a short summary of what was generated
• a summary of what was manually reviewed or modified
• explanations for important design decisions
• notes about limitations, assumptions, or areas requiring follow-up review

This additional context helps maintain:

• review quality
• long-term maintainability
• architectural consistency
• contributor trust

Reviewers may reject contributions that:

• cannot be reasonably explained
• introduce unclear abstractions
• add unnecessary complexity
• appear to contain unreviewed generated code
• conflict with ecosystem design principles

AI tools are intended to assist contributors, not replace engineering understanding or reviewer accountability.

---

Also working on an AGENTS.md - which should be read (?) by AI agents looking at the code where rules can be specified

[preetk25627-dotcom]

@eduardo-rodrigues @cantel
I've already opened a pull request (#1949) that implements the lightweight pull request template discussed here, focusing only on the AI usage disclosure checkbox and related fields, without introducing broader documentation changes. I'd appreciate any feedback on whether this aligns with the direction you had in mind for this issue.Thank you !

[cantel]

Hi all,

Just for awareness, I reviewed the PRs that were recently closed without merging (I had volunteered to take on this issue at the last HSF coordination meeting).

Reasons we closed PRs:

• PRs implement unnecessary changes.
• PRs implement unasked for changes (like addition of tools or web function we never discussed we needed).
• PR flooding of little changes (to note: the contributor was responsive, and after a conversation on 1 PR, they closed the rest with an apology)
• Accidental AI contribution in one case? PR deleted some text and didn’t match description of PR.

In the cases where they would have had time to respond, contributors were responsive and typically apologised.

We're just missing a CONTRIBUTING.md then?

I think these recent experiences can all be addressed with a CONTRIBUTING.md file covering basic etiquette, due diligence, small set of instructions and reasons why we will close a PR. There are several suggestions above and I'll get back with plans on CONTRIBUTING.md in a separate post (that perhaps @preetk25627-dotcom can help me refine).

About a pull-request-template PR

What are people's opinion on the pull-request-template, which was suggested in this issue and implemented as a PR by @preetk25627-dotcom (see above -thanks!) that asks for people to provide details on their AI-usage?

In my view, and since I don't think our recent experiences are due to AI usage issues directly, I feel a bit that it is not so useful clutter, and adds unnecessary drag.
Do we risk contributors not contributing something useful in that case?
And of course, in all cases we should be scrutinising PRs, whether that box is ticked or not.

I can see it from a "liability" point of view. In a very theoretical drastic scenario in which we merged something that has caused harm due to AI usage, we are liable for the merged code. A greater amount of liability can be shifted to the contributor if they did not close their use of AI when we asked them to. I think that aspect can be covered in CONTRIBUTING.md file as well: We state contributors should be transparent with their AI tool usage (the more detail the more useful) because they can be held liable if not(?).

Tagging @graeme-a-stewart , and perhaps @eduardo-rodrigues , who I think had differing opinions and can convince me otherwise :) It's definitely not a hill I'm going to die on, so I won't be dragging it on. Either way, I don't want to accept the PR without hearing from a couple of people first.

[preetk25627-dotcom]

@cantel
Thanks for the feedback and for taking the time to review this.

That makes sense to me. My intention with the PR was simply to help encourage transparency around AI usage, not to add unnecessary friction to the contribution process.

If the consensus is that this information would be better placed in a CONTRIBUTING.md file rather than a pull request template, I'm happy to update the PR accordingly or help draft and refine the CONTRIBUTING.md instead.

I'll wait for the discussion. Thanks !

[cantel]

Hello, given that we don't have a CONTRIBUTING file yet at all and so most of it would not be related to AI usage, I've created a new issue here to discuss a general one: #1951