GitHub Community
Allow us to block Copilot-generated issues (and PRs) from our own repositories #159749
Replies: 145 comments 225 replies
|
💬 Your Product Feedback Has Been Submitted 🎉 Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users. Here's what you can expect moving forward ⏩
Where to look to see what's shipping 👀
What you can do in the meantime 💻
As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities. Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐ |
Oh... Now here's where the lazy/fatigued people will use this to their advantage, these feautures don't only create Issues, but systemically close off and Ignoor older ones. It's one thing to be ghosted in need of help as a user of many projects (that you want to acknowledge as useful and share) but it's anoher to 'rub salt in the wounds' of personal issues that get closed by bot!!! Encouraging less human interaction is their M.O. IMO, SMH. (＃`Д´) |
|
It has always been possible to use a stale-bot to disrespect users by closing their issues automatically on a timer. It doesn't incorporate a Large Language Model. Nor does an automated form response involve a Large Language Model. I do agree with you that their M.O. is clearly to encourage less human interaction though. :) |
This comment was marked as off-topic.
This comment was marked as off-topic.
This comment was marked as off-topic.
This comment was marked as off-topic.
|
@corysimmons: pong 😉 |
|
gotta use copilot to create a bot to autodeny copilot |
im 99% sure they do, but managers are just hyped and they demand AI. |
|
peak |
|
Yeah @i-am-called-glitchy called it being a management issue. Our head of development wants us to go all-in on "AI" nonsense and I just want to smash my computer every time it's come up. |
(Myself) Speaking up for project users and not necessarily creators: Publically they (github) claim the main benefits but leave out the 'uno reverse' where the bots can close issues, pre-existing from it's system wide deployment, from users of the repo. I rather propper etiquette and communication than the owners/creators using your AI Messenger buddy to 'dump' your concerns/issues for you without a thought of a supporters feelings, this has happened in the same android/phone based repos I'm involved in, active use for potential OS porting/development (google free ofcourse). |
|
Shareholders hear about AI, become convinced that (insert product here) needs more AI features, then the managers tell the developers to make it. They have to make something related to AI, even if the developers know it's completely useless or actively harmful (and this "feature" is both). Ultimately the motivation is profit, not actually making a good product. Letting you block it would cause less people to use it, which means less impressive graphs to show off at the next meeting. Also, Codeberg is always an option, you get free unlimited storage/private repos, and avoid the constant corporate nonsense like this that plagues GitHub... |
|
I think there may be a editing error in your second full body paragraph- you suggest that the easiest way to waste everyone's time would be adding features to disable the waste of dev time that is Copilot PRs |
|
this seems to be fixed |
This comment was marked as off-topic.
This comment was marked as off-topic.
|
By the way, they can use all of your public (and possibly private) code you upload to GitHub to train their Copilot, regardless of your license restrictions, because this is stated in their Terms of Service! |
|
If you don't want your code to be read (by an AI or human or monkey), don't put it on GitHub. |
It's not about it being read, it's about it being included in large datasets without consent, and about AI being used to generate complete issues and PRs. Codeberg and Sourcehut respect people not wanting that, and so should MS |
|
I love the hate AI gets, just because you don't know how to use it doesn't mean you have to neglect what's never going away.. Feels like 1920 all over again but with AI... I dare anyone to give me any issue, any Protocol. I'll shit on it!! |
|
I want them to answer! Pleaseeee answer! |
|
I went to
And in my settings page under Copilot Features, I had an initial issue where the page rejected my attempts to set Anthropic Claude and Google Gemini to "Disabled". It appears to work now having closed the tab and opened a new one, but that was alarming to see... Still, two entries in that settings page are locked at Enabled, so it is impossible to disable "Copilot in GitHub.com" and "Copilot in GitHub Desktop", despite "Show Copilot" being disabled and helpfully hiding all the UI elements that had shown up. |
|
FYI, this only applies to PRs created by their agent (when you assign Copilot to an issue - only available through the more expensive Copilot Pro+ plan, not the tier that is provided to maintainers of popular projects and stuednt), not when a user creates a PR with the help of Copilot through prompts and PRs the finished work. Separately, for reviews (when you request a review from Copilot), |
|
Interesting, thanks. In my testing, I was able to block copilot-swe-agent but not copilot-pull-request-reviewer. I wonder if it's possible both bots are a special system construct, and some random github user coincidentally made an account named copilot-swe-agent on March 14. |
If so, it's a bug. Because when I click onto the username on logs, it redirects to the profile. |
|
Ah, yeah, I don't think a bug would do that! |
|
Sorry for the necro, at this point it seems copilot-pull-request-reviewer has been converted to an organization, thus not blockable. |
|
Please consider making this opt in, so that org and/or repository owners can have a choice. Or, if that's really not feasible, could there be an option to opt out? |
|
So you're telling me that JEP 369: Migrate to GitHub, which has been delivered was not about migrating to Git Hub? Cool story. They do use a dual-setup with bridges, which has now proven to have been an extremely smart choice. You can try to nit-pick but it won't change my actual argument: there's reasons why for every self-hosted top-project, there's 1,000 others on GH. You can read many of these reasons on the aforementioned JEP, which I'll link to again for your convenience: https://openjdk.org/jeps/369. |
|
Self-hosting is not an option for everyone. People do not self-host for the same reason that they do not grow their own food, build their own houses, or sew their own clothes: they do not have the time, resources or even skills to do this as well as professionals. |
That is correct, you can read all about how it was not about migrating to GitHub in the section:
"At this time" should be a dead giveaway. |
Cool story. Perhaps you may have noticed that I agree with you, and the current conversation is actually about whether self-hosting is a valid option for a minority group of professionals with funding? |
Right, but a lot of people here do have those skills, and supporting OSS with resources and skills that others can provide or pool is perhaps a more sustainable long term than people think. Eg, Codeberg for repository hosting, and sharing compute for runners for those that have those resources, donating to help fund where able etc, skills to help people setup. Currently sharing runners is a bit cumbersome, but that could be something Codeberg or others could change. Myself and other homelabbers I'm sure would be willing to donate compute, or skills. |
|
Legal advice often begs and commands that software is not exposed to the untested minefield that is AI. If a worker is not allowed to expose the code to AI... It'll be a hell of a day in the office if the entire org's is suddenly fair game. |
|
https://github.com/orgs/community/discussions/148571 Also see here for the legal concerns discussion. Unsurprisingly, Github hasn't responded yet. |
|
Here's the predictable result of this change: every single repository that a) is open, and b) is remotely large, is going to get DDOS'd by people submitting genAI issues against it. As a consequence, they will have to close the issue tracker, which will hurt GitHub. I understand that y'all are being told by your superiors to add genAI everywhere, and that you need to keep your jobs. Help us keep ours as well, by giving us the tools to stem the ongoing flood. |
|
I'd like a simple yes/no option to enable/disable CoPilot on GitHub. I am sure there are more refinements - eg: enable for A not for B etc but it's too intrusive in its current form. I think the idea of bots being able to report "issues" is ridiculous - but I really appreciate the Codeberg mention as I hadn't heard of them before. |
|
To add to @SirWumpus 's sidebar, I recently got self hosted Forgejo (Codeberg uses it) with my own runners and while the migration from Actions is a bit rough around the edges, I got them working in a few hours for a single project. The migration story will get better as more people hit the edges, and as long as you don't go in assuming your existing Actions will Just Work, the switch is very doable. Highly recommend developers at least have a go if you want off of GitHub. |
|
Gitea is now also a viable alternative, especially as it's (mostly) compatible with GH Actions, making the transition super smooth. Having a good 95% of all code hosted at a single provider is certainly not a good thing... |
|
@kblaschke , Forgejo/Codeberg is a fork of Gitea. I am not qualified to evaluate the differences between them technically, but if you look closely at the Gitea web page you will realize Gitea may not be the #1 choice if you're trying to avoid services that use generative AI. |
|
The technical differences are (at least for now) very minimal. Forgejo came into existence due to differences in views about governance. They later became a hard fork because of rising tensions. Forgejo is definitely very nice to work with though! |
|
@RivenSkaye when MS bought GH into their butcherhouse, the difference in this and the alternates were very minimal.... strange?? |
|
EDIT: The suggestion to use Copilot, seen on the screenshot below, has been removed. I'm not removing it from this issue because keeping historical context is important, but I also don't want people thinking this is still true, so I've added this edit note before my comment :P I just noticed that it's part of the standard "Create new issue" page. I can't wait for people to start creating issues that aren't even real because AI hallucinated something from the description given by the user :/ Please, GitHub, allow us to hide these from our issue trackers at least! |
|
Don't worry, I tend to write walls of text too 😄 Usually not sectioned, I guess, but I imagine you probably wouldn't be classed as AI. Current AI models imo just have a certain style (linguisticaly too, not just form-wise) of writing things that you are unlikely to replicate by coincidence. |
|
Oh, beautiful. Using that will also happily ignore the issue forms on your repository that you've spent ages on creating so that users will actually give you the info that you need 🤡 Oh FFS! |
|
@JackAtOmenApps @JackAtOmenApps @foosel FYI
|
|
Yeah, one of those maintainers was me. |
|
Thanks for that ❤️ |
I tried creating an issue on my own repo through Copilot and I ended up as its author, not Copilot, so this might not work. |
|
When you did this, was the comment history of you querying the bot— or any other paper trail which would allow the repo maintainer to detect that you used the copilot feature to create your submission— left behind? |
|
@mcclure, the issue can be seen here: Cog-Creators/Red-DiscordBot#6575 I can't find any other visible indication of the fact that it was generated in either the UI or API (https://api.github.com/repos/Cog-Creators/Red-DiscordBot/issues/6575), but feel free to check for yourself. |
|
Thank you, so that's extremely alarming. Not only does Github provide site-integrated tools that allow users to submit fake, machine-generated issues and PRs in violation of my project's submission rules, they then cover up the fact the tools were used in the first place, preventing me from enforcing my project's rules manually. This increases the maintainer load of detecting and removing this content even further. I'm going to have to basically introduce a verbal CAPTCHA. |
|
Some organizations have policies preventing use of generative AI, especially while concerns around rights are being debated in countries around the world. Making this feature opt-in would give those organizations a way to avoid someone inadvertently getting in hot water with their boss because they accidentally accept a PR. |
|
As acdha says making copilot issue/pr support opt in would be infinitely better than making it opt out. However Microsoft has declined many previous requests to allow even opt-outs on other parts of the copilot feature, and both GitHub and VSCode have at one time or another added copilot disable buttons which later copilot interface nubs simply ignore. So I just assumed, when submitting the feature request, that they would not even consider something so pro-user as making their planet-melter tech opt-in. |
|
https://github.com/orgs/community/discussions/148571 Also see here for the legal concerns discussion around co-pilot, especially the questionable training data. Unsurprisingly, Github hasn't responded yet. |
|
There is already a huge problem with low-effort AI-hallucinated issues that plague FOSS projects, that are costing the maintainers a lot of time to filter through to the point of burnout. This will only make it worse. Enabling and even actively pushing users through aggresive marketing and pop-ups towards doing exactly that is extremely hostile UX change for maintainers, to the point where projects will have to consider either closing issues or flat out moving away from this platform, which at this point I absolutely support. |
I agree, As stated previously, a somewhat formal conversation seems to be dwindling (by design IMO) using Ai as a MITM pimp to reject or accept issues as they are experienced. Power is given to the bot to close Issues that aren't even solved. Very dangerous for serious security or privacy functions in any project as you stated, even potential censorship in extreme cases. Not too far from impossible to manipulate human interaction before completely replacing it. |
As much as I agree with the first the other points, AI safeguards are often excessive and don't stop people from using an uncensored model. You can take illegal pictures with a camera, hack someone with a Bash shell and hit someone with a hammer, why put safeguards on AI and not those? |
|
It is possible to ethically use:
|
Because a hammer doesn't represent a credible existential-threat to humanity. Check out Roman Yampolskiy's interviews or https://gstonepress.carrd.co/ for some of the many experts extremely-worried about AI safe-guarding. |
|
I think this Github issue would very quickly turn into a ideological joust if we let conversations derive that much from the original point. It'd offer many anchor points for AI-apologists to hang onto and/or make inaccurate comparisons or call people Luddites. It may also be less convincing to the people taking decisions at Github, since they are clearly convinced they know what their customers need better than the customer themselves. Instead, I recommend sticking to the very clear, unalterable, and inarguable facts: outside of any ideology, slippery slope theory or abstract morals, AI-generated code and issues are spam and trash which make life harder for everyone, costlier, at no benefit. I feel like harping on this and this only is more likely to resonate with everyone, no matter if they're a techbro or not. (not that I have much hope it can change anything, there's too much money involved and the decision takers clearly need to justify this investment somehow with so-called innovation, even if it is a pure negative -- what matters is the marketing spin they can put on it. But we should try nonetheless!) |
|
I don't think I have the repeat again what has been said previously or in the original post, but I will. Pushing this on OSS contributors is one of the first worst thing that Github could do. This should be opt-in not opt-out. |
|
It's the second thing, actually. |
|
The issue here is it's not even opt-out (not that that isn't a problem too of course) |
|
We would have to consider pulling all of our projects from GH if this becomes a problem and there are no tools to address it. I hope GH and Microsoft can get into their senses that AI will not solve our problems, especially when we don't want it to. We spent 4 years building a good contributor culture in our projects and we'd love it to continue without forcing AI down our throats. |
|
First Google Code, now Github... not the first, not the last. |
|
Honestly, I think I'm just going to leave github and replace any of my active projects with a readme containing a link |
|
@alicealysia, this is a good approach. Check out an example of people doing this: https://github.com/conservancy/npo-ledger-cli. |
|
Thank you, also linked to some alternative resources. Wound up paying sourcehut a bunch because it's nice to just, have a place to dump my code. |
|
This is a really bad idea. It is guaranteed to generate inordinate amounts of spam, that project owners are going to have to deal with, and subsequently - to projects and orgs leaving for the competition. Please don't. |
|
A new blog post has been published yesterday: How to create issues and pull requests in record time on GitHub. GitHub could not make it any clearer about what they think of the comments above. Of interest is this question from their FAQ:
So any misuse of the feature is none of GitHub business. What could possibly go wrong ? |
|
Have to agree... at this point I'm pretty convinced that a year or two down the line, everybody will point at Copilot as the thing that caused the downfall of Github. I'm here to work on code, not to entertain punters and their toys. I've seen and used GenAI enough to know that it's bad news for anyone who takes some pride in their work and cares about quality. |
This constitutes a proposal to implement a tradeoff between:
I don't have any doubt whatsoever about the effects of this. The former is negligible. This is going to drastically increase the quantity of LLM submitted issues. I don't see how you can possibly claim that you're going to reduce spam by giving the vast majority of people tools they wouldn't otherwise have for creating spam, and politely asking them to review the output before submitting. It's not really clear to me what constitutes "review" anyway. Users who submit such issues are going to see details they didn't input into the LLM and think that that's the LLM doing what it's "supposed to do". i.e. filling in the blanks based on known facts about what the software is supposed to do, and perhaps even doing a test run and analyzing the output. Obviously, if they think that's what the LLM is supposed to do, they are wrong because they haven't taken study courses on how LLMs work. All the LLM can do is "rephrase, in your own words, [...]" assuming the LLM doesn't also add in a hallucination. Users who used GitHub tools in good faith, believing that it was the recommended way to do things, are going to find themselves permanently banned from projects with NO appeal, when it is discovered that they are "lying" to project developers.
Having a way to opt out will at least let project developers who are following this news, set their expectations in advance, and avoid this kind of "you lied to me, be banned forever" interaction. |
|
If the LLMs were so good, everybody would want to use them, instead of having them shoved down our throats everywhere. |
|
@Vxrpenter I'm not worried. An Analogy: Remember Digg? Digg died within weeks when they did some changes that alientated the majority of its code base. Within weeks Digg was bleeding out terminally and Reddit suddenly became the thing. Nobody remembers Digg. But Digg was the old Reddit. OLD reddit. See also MySpace and facebook. Things can change fast. |
Yes, it is what it is, we need to go back to self-hosting. Cooperates don't have our best in mind. |
|
I thought GitHub employees were supposed to be knowledgeable about how things like LLMs work at the most basic level... It's a high-tech next-word predictor. It's a statistical model. Not a brain. And a brain is required to coherently report an issue in such a way that a developer could fix it. It requires an inner understanding of what the heck you're dealing with, not just knowing a general idea of what you're supposed to talk about - which is the only thing LLMs "know". Moreover, as so many others have mentioned, it's only going to result in an uptick of spam. Stop it with this overhyped, resource wasting BULLS***. It's not helping anyone. |
|
I get existential anxiety about AI and not being able to completely avoid it, so I can make the argument that I have to avoid AI for health reasons. I can't get a new flagship phone because one of the major selling points is an anxiety machine that could possibly be tightly integrated into the OS. I want to talk to real people because that makes me feel more connected and brings a sense of fulfillment. I don't want to talk to LLMs because AI is just a program that does a bunch of math and doesn't have feelings. The only time I had a genuine use for an LLM was to find a specific website that just so happened to be down for over a year. Please don't force me to use your sledgehammer when all I need is a screwdriver!!! Please. |
|
I abandoned Windows when they [Microsoft] pushed Recall. I will abandon Github for this nonsense as well. Thanks to previous commenters for listing some alternatives; I'll be looking into those. I'm not a major loss by any means - my contributions are narrow, most of my projects are from my student days. That just means I'll be in the first wave of rats to leave the ship. Cheers. |
|
It seems they would rather try to trick me into buying Copilot Pro+ by adding weird popups: |
|
come on down to codeberg and have yo self a time |
|
@Lampe2020 One option is using GitHub as a read-only mirror of your repo, and then have everything else (issue tracker, pages, etc.) be on Codeberg. Also I know you haven't mentioned it, but surprisingly they have an equivalent to GitHub Actions as well. |
|
@Anonymous941 that is what I do. However, Github doesn't allow disabling security vulnerability submission forms, even if you don't accept nor read reports on Github, and also doesn't allow disabling pull request. Meanwhile, the obvious way to discuss security related questions for small projects, private filed issues that aren't limited to just known actual vulnerabilities, still haven't been implemented. My legit assumption is that it didn't look as flashy to big money users as some sort of CVE registry that can't be disabled. |
|
@ell1e Maybe you can do what Linux does, and use a bot which replies to pull requests and closes them? That's not a perfect solution of course |
|
I do that actually, with github workflows, which means I'm wasting github's compute for this. But that's an absurd fix to an absurd problem that shouldn't exist. |
|
@ell1e I agree, clearly Microsoft no longer cares about making a functional website |
|
This issue has been given coverage on a technology news website. This very bad press was my first impression of the copilot issues feature. It's not looking good at all. How would microsoft even make money from this junk? The article in question: |
|
Possibly relevant as well: https://github.com/orgs/community/discussions/148571 |
It's not about making money from users, it's about impressing investors with 583 new AI products so they buy more stock |
|
I have, a little while back, started actively moving things over. I've also posted about my distaste on my blog and think the recent media coverage is a good turning point to more seriously consider writing an open letter to ask these people to just stop this behavior. |
|
First: I support the position of most here, that this should be an opt-in only "feature" and disabled by default, with clear rules and user controls for disabling it. Microsoft are being jerks. But second, I wonder: if one gives an AI agency, via a web browser, via an IDE plugin, via its own capabilities, whatever, what's to stop it acting as a human and submitting Issues, Features, even PRs at will through the web UI? Seems to me that like it or not, if Issue/Feature Req reporting and PRs are open to non-members to a project, eventually we're going to have AIs submitting these things, desired or not, and tagged as AI or not, at their own discretion. And the next knot in the rope: If IA( thinks it)'s so smart, why doesn't it just fork a project and do it itself, locking out outsiders? |
|
It already exists, it's called "computer use". But I'm confused, how does this have to do with Copilot? |
What? |
You're right, my post was kind of confusing, sorry about that.
I was pointing out that it's not hypothetical, because computer use is a feature that lets an LLM view your screen and click/type (a disaster waiting to happen IMO, but that's beside the point). It's not exactly what was described, but it's the same difference. To be fair to @jimmiedave, it's not that unrealistic of an idea, just very situational.
People could already do this by just typing into any LLM "write a bug report for project X because it says this error when I try to open it" and copy-paste it. The issue is that when it doesn't even require leaving GitHub and is literally a mouse click away, 500x more people are going to do it, and showing it right under the maintainers' templates tricks people into thinking that it's acceptible. |
|
The abuse of AI scrapers has really called into question whether things like robots.txt and licenses were ever truly respected online period. As in, who was the "open web" really open for in the first place? Bouncing off that thought, y'all should assume that Microsoft and other firms are going to go full throttle into AI (unless they come in here and say otherwise). That means that if you don't want your code scraped, you need to adjust your threat model accordingly. And remember the bystander effect. Don't expect "someone else" to magically fix this. The time for you to act is now. |
|
For the code scraping issue, feel free to upvote and weigh in here. |
This comment was marked as spam.
This comment was marked as spam.
|
Clearly an AI-generated ad, someone delete this spam please |
|
Hello there, GitHub Universe Team! 🌌🚀👩💻👨💻 First and foremost, allow me to preface this remarkably humble yet devastatingly thorough transmission with a deeply heartfelt, profoundly genuine, and elaborately verbose expression of my enduring gratitude for the continuous, relentless, unstoppable innovation you all tirelessly bring to the world of collaborative version-controlled software engineering development synergy in the ever-expanding digital galaxy of tomorrow. 💖✨🌍📦 Now, let us gently pivot into the infinitely nuanced subject of concern that has caused a measurable tremor in the otherwise relatively stable tectonic plate that is my software maintenance workflow ecosystem. 🌋💼 I was, as I often am, perusing the sprawling and scintillating sea of changelog entries on GitHub’s Blog™—a digital cornucopia of productivity-centered announcements—when I stumbled, tripped, fell, rolled downhill, and splashed face-first into the following two particular hyperlinks of utmost relevance: 🔗 https://github.blog/changelog/2025-05-19-creating-issues-with-copilot-on-github-com-is-in-public-preview/ Upon reading these momentous URLs and the glowing prose within them, I was struck—nay, utterly flabbergasted—by a chilling realization that pierced through my soul like a rogue semicolon in production: namely, the dawning awareness that GitHub is now actively enabling, encouraging, and dare I say celebrating the automated, AI-generated submission of Issues™ and Pull Requests™—a development that I must now unequivocally label as "epistemologically catastrophic" for human-centric open-source stewardship. 🚨👾 Let me be maximally clear and undeniably explicit: I, a carbon-based, keyboard-using, emotionally burdened human maintainer of several repositories which I have lovingly curated and debugged during many late evenings with nothing but coffee and existential dread to sustain me, do not want, accept, appreciate, enjoy, tolerate, respect, endorse, support, condone, encourage, request, or even contemplate the presence of AI-generated issues or pull requests within the sacred code-temples that are my GitHub repositories. 🛑 Indeed, the mere thought of receiving a Copilot-generated issue or PR fills me with the same existential nausea one might associate with discovering a merge conflict in the To put it in a delightfully redundant and unrelentingly detailed fashion: these bot-born communications are not "helpful." They are not "time-saving." They are not "innovative." They are, from my point of view, a repetitive and contextually incoherent deluge of machine-produced techno-sludge. 🧠⚙️💩 As a direct result of this deeply troubling development, I would like to formally submit a proposal, nay, a plea, nay, a desperate cry into the heartless abyss of automated contribution systems, for the following feature: 🧩 Please provide—immediately if not sooner—a per-account and/or per-repository setting, checkbox, toggle, switch, lever, slider, dropdown, checkbox again (just to be sure), or a vintage 19th-century brass valve with hissing steam, that allows repository owners and maintainers to BLOCK, PREVENT, BANISH, EXCLUDE, YEET, NIX, or otherwise TERMINATE WITH EXTREME PREJUDICE the use of GitHub Copilot, its bots, agents, assistants, interns, scribes, scribblers, and miscellaneous AI gremlins from contributing to or interacting with our repositories in any form, shape, gesture, echo, vibe, or spiritual wavelength. 🔒🤖🚫 Furthermore—and I say this with no small degree of verbosity—the "copilot" bot account, which you appear to have blessed with supernatural immunity from normal user blocking protocols (possibly in accordance with some sort of ancient arcane GitHub Constitution I have not been permitted to read), must be made blockable. Not just metaphorically. I mean literally blockable. As in: I click Block. It gets blocked. The end. No exceptions. No special cases. No pop-ups saying “you can’t do that.” Just the sweet, sweet sound of silence as the bot is digitally escorted off my repo like a heckler at a live coding session. 🎤🔇 🧠📉 Because without that—without this tool, this checkbox, this guardian against the endless tide of synthetic spam—I may be compelled, against my will and better judgment, to do what no responsible open source maintainer should have to consider: shutting down my project’s issue tracker entirely. Turning off PRs. Pulling the plug. Going dark. Going ghost. Going full hermit-coder-mode. Relocating to Codeberg. Scribbling issues on a napkin and mailing them by pigeon. Who knows? The possibilities are boundless, and most of them are bad for GitHub. 📉📉📉 As of today, my projects’ codes of conduct are being updated to formally and explicitly prohibit AI-generated contributions—issues, PRs, or otherwise. This may not stop the bots, but it will give me moral clarity when I immediately close, laugh at, or ceremoniously delete such submissions from existence like a digital Thanos. So I close this epically verbose feedback message with a request, no—a mission. Give us the power to protect our spaces from the dystopian influx of AI-generated nonsense. Allow us to say, with dignity, "No, Copilot. Not here. Not today. Not ever." 🛑⚔️🚷 Thank you for coming to my TED Talk. I’ll be here all week, manually triaging issues written by actual humans. 👋 ¹ Footnote: Disclaimer: This message was generated with T9 predictive text technology on my Nokia phone.
|
|
I think you are likely to get better issues if a bot creates them vs a human. How many of you have set up an issue template only for users to completely ignore it? How often has a human filed a bug that was missing or had completely wrong information? Honestly, people should try it and experience it before asking for features to remove it. |
|
Not filling templates doesn't happen that often; when it happens, it doesn't matter that much. If you follow a large open source project closely, you'll find that most people actually do use the template. There are also people and/or systems who's job it is to herd and triage issues. Now the second part; as for trying out bot suggestions, there are plenty of examples out there; the reason people are complaining is specifically because they've seen the results. Why do you presume people are reacting? |
|
Speaking as an actual developer for multiple popular projects that get tens of thousands of issue reports, I suppose I may count as an actual authority figure on the subject. I'm quite familiar with issues that have to be closed due to:
I will take a million bugs like this (exceedingly easy to analyze and determine that they are invalid or NEEDINFO) over a single LLM-generated "very thoroughly filled out bug report that sounds convincing but turned out to be gaslighting me". The two are apples and oranges. The dividing line is that LLMs are malicious, abusive, and result in the human that uses the LLM being a liar. I do not care why a human user who interacts with me is a liar and is trying to gaslight me. If I catch a human user doing this, my natural instinct is to permanently ban that human user from my community. For what are hopefully obvious reasons, the appeals process for "banned from the community for lying and gaslighting others" is... complicated. How do you prove that you're telling the truth and not gaslighting people when you try to tell them that you've changed your behavior and no longer go around lying to people and gaslighting them? But, the reason why this matters with regard to LLMs is that I've never had someone do that to me, on their own. LLMs, however, cause people1 to do this by default. I'll also state up front that I take a pretty dim view of people trying to gaslight me into thinking that it's actually in my best interest to accept gaslighting as normal and acceptable behavior. Footnotes
|
|
And it is for the sake of users, moreso than for the sake of myself, that I would like to block Copilot. It's probably not pleasant for users to discover that they are malicious liars who gaslight people, and get banned from the communities they tried to interact with. Removing the temptation to do so in the first place is a good way to cool the anger before it gets started. |
|
Until we reach the point where LLMs are controllable tech with deterministic outcomes, it's impossible to trust what parts of its output are hallucinated garbage. It takes effort to find that out, and besides being an earth scorching tech with no tangible benefit in this are, it wastes a lot of human time to figure out what (if anything) is even true. It's being deployed like this to push up numbers for investors, with no regard for the havoc it wreaks. Imagine deployments like these happening in schools, where LLMs wiuld hallucinate half the class materials and answers. It'd be a global outrage, because people have seen memes about how current gen LLMs react to questions like "how many times is the letter E in the word 'berate'?" |
|
Surely people random-generating issues for a repository and actively creating misinformation is not actually better than someone just providing missing info. |
|
If GitHub won't provide a way to opt out of other people's AI-generated issues and code, you can always try making a This gets picked up automatically as a repository-wide instruction set and basically tells Copilot to refuse everything with a "nothing else matters" for added weight, hopefully rendering it useless. |
|
Absolutely. Asking for an utmost-basic explanation in the last couple-days of September for a discussion that started May is crazy. |
|
Why would it be impossible to introduce a per-repository setting (with per-user/organisationd defaults) such that if set and somebody mentiones that project on https://github.com/copilot, instead of actually processing the prompt, auto-reply with a message like
? 😉 |
|
@MisinformedDNA You continue to defend AI as if you had a personal emotional stake in it, to the point of being blind to both the original opening post, and refusing to accept, despite having been told, that many people not wanting this use AI daily, know AI well, are sometimes AI researchers, and still don't want Github to push this feature. This is unproductive, because even if you couch your words in polite language, they still are an ad hominem; you're essentially accusing people of being against something because they don't know enough. Instead of engaging with the argument, you're dismissing it on the basis of perceived lack of knowledge, character, or motive of the people speaking (whatsmore, without even asking about them). But it's even more unproductive than that; If it was just ad hominems and not respecting what people say, it'd be one thing; but it's worse because you do not care at all about the point people are making, and only want them to say AI is "good" somehow. It's just taking the conversation astray, detracting from the point, just to satisfy some kind of personal ego/pet peeve thing. Just completely off subject. No one here cares if you like AI or not, or if they should like AI or not. When you feign to go back to the discussion, you still do not engage with it with any level of honesty or integrity. Of course, making things "easy" is the point. You speak as if making things "easy" is meaningless. If it is "easy" to travel from Paris to New York because we have planes instead of horse, people do it more. Making things "easy" means people do something more, is the entire point. This shouldn't need explanations. There's no need to have "AI" fix it, Github can just not do the thing. I suggest you please engage with the argument and stop moving the conversation elsewhere, to battle imaginary windmills. Or, if you are unable to engage rationally, simply do not engage with the conversation at all; you certainly have better to do with your time. |
Besides-that much-more has been-offered in this discussion since it-started back in May. Him asking how is done entirely in bad-faith, either deliberately or because he's unable to do other-wise. There's no point in engaging any-more, he's also a wind-mill. |
|
What a funny workaround |
and others
Select Topic Area
Product Feedback
Feature Area
Issues
Body
I find the following two news items on the front page:
https://github.blog/changelog/2025-05-19-creating-issues-with-copilot-on-github-com-is-in-public-preview/
https://github.blog/changelog/2025-05-19-github-copilot-coding-agent-in-public-preview/
This says to me that github will soon start allowing github users to submit issues which they did not write themselves and were machine-generated. I would consider these issues/PRs to be both a waste of my time and a violation of my projects' code of conduct¹. Filtering out AI-generated issues/PRs will become an additional burden for me as a maintainer, wasting not only my time, but also the time of the issue submitters (who generated "AI" content I will not respond to), as well as the time of your server (which had to prepare a response I will close without response).
As I am not the only person on this website with "AI"-hostile beliefs, the most straightforward way to avoid wasting a lot of effort by literally everyone is if Github allowed accounts/repositories to have a checkbox or something blocking use of built-in Copilot tools on designated repos/all repos on the account. If we are not granted these tools, and "AI" junk submissions become a problem, I may be forced to take drastic actions such as closing issues and PRs on my repos entirely, and moving issue hosting to sites such as Codeberg which do not have these maintainer-hostile tools built directly into the website.
Note: Because it appears that both issues and PRs written this way are posted by the "copilot" bot, a straightforward way to implement this would be if users could simply block the "copilot" bot. In my testing, it appears that you have special-cased "copilot" so that it is exempt from the block feature.
So you could satisfy my feature request by just not doing that.
¹ i don't at this time have codes of conduct on all my projects, but i will now be adding them for purposes of barring "AI"-generated submissions
Guidelines
All reactions