Re: [gwt-contrib] Digest for google-web-toolkit-contributors@googlegroups.com - 1 update in 1 topic

[Tim Macpherson]

Not a channel for general chat but just wondering if anyone wants to suggest whether LLM generated pull requests with purely LLM follow up will ever be acceptable

Yahoo Mail: Search, Organize, Conquer

On Thu, Jul 23, 2026 at 12:50 PM, google-web-tool...@googlegroups.com

<google-web-tool...@googlegroups.com> wrote:

google-web-tool...@googlegroups.com

Google Groups

Topic digest
View all topics

• PR template change to require an issue first? - 1 Update

PR template change to require an issue first?

Jens <jens.ne...@gmail.com>: Jul 23 03:54AM -0700 Generally fine, although I would make it more strict and directly say that the PR will be closed right away. A reviewer should not be required to spend extra time to decide if the PR is good enough to keep open and wait for an issue.   But given that these kind of pull requests are usually AI driven we might want to think about an AI-Tool usage policy. I already posted it in chat but I kind of like the writeup done by LLVM: https://llvm.org/docs/AIToolPolicy.html   Such a policy can then also contain information about creating an issue first. The LLVM policy basically allows AI tool use (with approriate commit/PR trailer to indicate tool use) but requires that a human fully understands the code in question and communicates with the reviewer personally. Fully automated tools with automated comment responses or copy/pasted AI responses are disallowed. We already had a few PRs like these.   -- J.     Colin Alworth schrieb am Montag, 20. Juli 2026 um 14:54:55 UTC+2:

Back to top

You received this digest because you're subscribed to updates for this group. You can change your settings on the group membership page. To unsubscribe from this group and stop receiving emails from it send an email to google-web-toolkit-co...@googlegroups.com.

[Colin Alworth]

I'm not a lawyer, so I can't authoritatively speak on this, so this is just my understanding. I'm generally going to stick to the conservative side here, so that even if something is likely legal but untested, we should draw our metaphorical line just a little closer to "definitely legal, widely accepted".

In several jurisdictions, purely LLM-created content has been deemed not copyrightable, not creative work, so as you don't own it, you can't license it as Apache (or anything else). Our policy is to be Apache v2 licensed, so adding other content to the project could cause more headaches than it is worth. So no, purely LLM-created content probably can't be permitted.

Another area we need to take care is in our JRE emulation - it is my understanding that we should not accept LLM contributions in this area, since it runs a fairly high risk of reproducing code from OpenJDK or other non-apache licensed JRE implementations. Outside of that, it is my personal belief that LLM usage in moderation may be safe to accept.

I think there is a lot of good prior art in conservative LLM policies out there that we should draw on when building our own policy. I think that the above constraints (subject to my own, possibly flawed, interpretation of the law where I reside) are clear from existing policy, but stating them outright might help to have specific guidelines to point to.