Stricter anti-AI Policy

[poyribn]

Making an anti‑AI Policy for Godot is definitely the correct Choice.
But in my Opinion, it is still too relaxed for how much Damage LLMs really do to our Planet Earth.
LLM‑generated Code should be completely forbidden from entering the Main Branch in the Future.

• LLMs plagiarize Code, potentially causing legal Troubles over Source Code Licensing.

• LLM Compute Centers consume massive Amounts of Water, making many Organisms die from Dehydration.

• LLM Compute Centers produce much toxic Waste Gas, making it hazardous to breathe for anyone living nearby.

• LLM Vendors won’t stop buying out Computer Parts for as long as End Users use LLMs.
  This even makes crappy Computers cost as much as a good Computer should cost.
  This means i can’t make a Game with cool, even just moderately compute‑intensive Systems and Graphics, and expect potential Players to have a Computer that can play it.

• LLM Outputs are not always Computer-generated.
  Because LLM Vendors know their Computer Output is often inaccurate,
  they also employ Slaves in third-World Countries to answer Queries accurately.
  This is especially true for War‑related Queries in military AI.

There should probably also be a GitHub Issue Template to report LLM-generated Contributions in Case they make it into the Source Code undetected,
or they have made it into the Source Code before the new Policy removed all Tolerance for generative AI.

[scarzehd]

While I agree with your sentiment, I'm not sure it's possible to enforce. Unchecked LLM outputs are relatively easy to detect since they tend to be of low quality, but how can we know if an experienced developer copied a function or two out of ChatGPT?

As for existing AI-generated code in the codebase, what good does it do us to have a human rewrite it? Maybe it displays our commitment to the policy, but any harm to the environment and to the economy has already been done. As long as the code works, I don't think there's a reason to spend a bunch of time rewriting it to no benefit to the engine or its users.

[poyribn]

Yes, it is impossible to enforce completely
but the harder we try, the less Damage we do.
That’s the actual Point of a strict anti‑AI Policy.

[Naros]

LLMs plagiarize Code, potentially causing legal Troubles over Source Code Licensing.

This issue exists even without LLMs. Nothing prevents a random contributor from cut-and-pasting code from another project, failing to attribute it, and passing it off as their own, which can put the project or YOU in the same situation.

The way OSS projects avoid liability is by requiring a contribution license agreement to be signed upfront or requiring the contributor to sign off all commits to adhere to a developer certificate of origin, which, in most OSS legal terms, binds YOU, the contributor, to bear all liability for a contribution that was made with false pretenses.

While the economic, social, environmental, and ethical concerns are valid in their own right, it's generally best for OSS projects not to get involved in these terms because they can alienate a subset of the project's community who may have a different opinion. So these really should play 0 part in Godot's AI policy regulation imo.

[sockeye-d]

This issue exists even without LLMs. Nothing prevents a random contributor from cut-and-pasting code from another project, failing to attribute it, and passing it off as their own, which can put the project or YOU in the same situation.

That's like saying solar power has a environmental cost involved in manufacturing so we should just stick with coal-fueled power plants because they both have their own respective cost.

they can alienate a subset of the project's community who may have a different opinion

Maybe those aren't people you want contributions from anyway?

[Naros]

I think my point got read as "plagiarism happens either way, so the risk is the same", and that's not what I am arguing.

I'm not claiming the two sources carry equal risk. I'm saying a project's protection against uncredited code is the same regardless of where it comes from. A DCO sign-off or CLA makes the contributor legally responsible for asserting they have the right to submit the change. That holds whether they pasted from another repository or from an AI model. A ban on LLM output doesn't add any protection the sign-off doesn't already provide; it just adds a rule that can't always be verified at review time. The sign-off is the enforceable part.

So I'd put the question back: what does a ban protect against that the DCO doesn't? If the answer is "the volume of plagiarized code goes up," then the fix is stronger review and attribution requirements, which help against both sources, not a rule that applies only to one.

On the second point, projects that take positions on contested issues outside their scope tend to lose contributors on both sides, not just the ones you'd be happy to see go. Keeping the policy focused on what the project can actually enforce, licensing and code quality, lets people with different viewpoints keep working together on the same codebase toward common goals.

Of course, YMMV.

[mrjustaguy]

Making an anti‑AI Policy for Godot is definitely the correct Choice. But in my Opinion, it is still too relaxed for how much Damage LLMs really do to our Planet Earth. LLM‑generated Code should be completely forbidden from entering the Main Branch in the Future.

• LLMs plagiarize Code, potentially causing legal Troubles over Source Code Licensing.
• LLM Compute Centers consume massive Amounts of Water, making many Organisms die from Dehydration.
• LLM Compute Centers produce much toxic Waste Gas, making it hazardous to breathe for anyone living nearby.
• LLM Vendors won’t stop buying out Computer Parts for as long as End Users use LLMs.
  This even makes crappy Computers cost as much as a good Computer should cost.
  This means i can’t make a Game with cool, even just moderately compute‑intensive Systems and Graphics, and expect potential Players to have a Computer that can play it.
• LLM Outputs are not always Computer-generated.
  Because LLM Vendors know their Computer Output is often inaccurate,
  they also employ Slaves in third-World Countries to answer Queries accurately.
  This is especially true for War‑related Queries in military AI.

There should probably also be a GitHub Issue Template to report LLM-generated Contributions in Case they make it into the Source Code undetected, or they have made it into the Source Code before the new Policy removed all Tolerance for generative AI.

Much of this information is Wrong.

1. Plagirization doesn't require LLMs to happen in OSS, so this isn't a new problem caused by AI. Likewise the Courts have yet to Decide if Gen AI counts as such or not, and to which extents.
2. Water usage of Data centers (even AI ones) is a drop in the bucket. Further to that we Live on a Planet that's covered in the stuff, and the real issue is Desalination efficiency, which AI is itself a tool actively used to improve Desalination Tech (among a lot of other tech)
   Energy usage is also a Drop in the Bucket when you see global power consumption for everything combined even if you filter out AI heavy countries https://en.wikipedia.org/wiki/Electric_energy_consumption
3. Computers do not produce Toxic Waste Gas besides their Production or Possible exotic Coolant leaks
4. Even if the entire world Boycotts AI use, they will still be building and maintaining their Data Centers, they have ample uses for them Beyond End User AI.
5. Computer Hardware Cost has not increased substantially overall. It's not in it's lowest state, but it was FAR worse back around Covid. The only Components that are substantially impacted are Memory and SSDs, and it only looks so grim given how cheap these things have gotten as we were in one of the Memory and Flash Industry's Cyclical Dips where Memory and Flash get Dirt Cheap.
   Likewise looking locally right now, buying a Decent Computer Now is like 1-1.3k Euro for a Decent Hexacore (i5 14400f or r5 5600x), with 16gb of memory, 1tb of SSD storage, and a Decent GPU (like an RX 7600 or RTX 5060) prebuilt. For reference I got an i3 10105f, and an RX 6600, and I had built it myself. Took some 300€ on the CPU side, and 240€ on the GPU (note this price was very short lived and would've expected more like 300€ typically), and I had Windows, Storage and Case and all already, and paid no Prebuild Tax. So let's say the PC was equivalent to a 900€-1.1k€ prebuilt one at the time
   Building a DYI PC full right now I'd get 16gb of ddr4 for 140€, a r5 5600 for 120€, and a MBO for 60€, for like 320€ on the CPU side, and an RX 9060xt 8gb for 400€ and that's only because the shop I'm looking at doesn't have anything else right now below that. All of the components and Prebuilt PCs are available in stock at a fairly small Local retailer that's a 20 minute walk away from me. Larger retailers have some slightly better deals if you shop around. This means in effect there has been a 20-30% increase in PC prices for the typical buyer, tho Higher end buyers are much more heavily impacted.
6. How the hell can an LLM output not be Computer Generated!? What BS is this? No Human would be able to answer queries as fast, even if you threw accuracy completely out the window. The fastest typist in the World is able to get around 170 wpm for an hour https://en.wikipedia.org/wiki/Barbara_Blackburn_(typist)
   AI for context is able to within 2 seconds generate over 1 thousand characters for a guestimated 50-100 words per second if you divide that by say 10 characters for each word

Also some additional Context, I've been following Machine Learning Back before it was beating Pros in Go, SC2 and before ChatGPT started this whole LLM/Gen AI Craze, and I'm neither very Pro AI nor very Anti-AI, and in fact I have more to be Anti-AI then Pro AI given it has basically Made a Decade of Work amount to more or less: "Cool, you can throw it in the trash and get a REAL job." and It's utilities to me are either just some cool little toys to play with that I wouldn't really miss or replacing me in the things I typically enjoy.
Others however have gained quite a bit with this tech, now that it's somewhat mature (out of 5 people in my household, 3 are dayly using AI and it's genuinely being helpful to all 3 of them, and only 2 of us don't have anything very meaningful to do with it and thus aren't using it, right now atleast. I'm also excluding AI search as that's basically unavoidable at this point, and the summaries are still only really consistently useful to skim through to see if it understood what you're looking for and then grab the links it found)

[mrjustaguy]

1. You underestimate how much people copy other people, especially in the Programming world.
2. AI has been used for various branches of Research (such as Material Discovery) for many, many years now, which are themselves contributing to improving desalination tech, and many, MANY other technologies, many of which you are likely using right now and do not realize it.
3. Paywalled, not reading that
4. Clearly you don't know....

5. RAM is but one of many components in a Computer. RAM may have increased in price (from 40-50€ (in like 2023 or sth) for 16gb ddr4 3200MT/s to 140-150€ today, here locally) and Yes I can with a straight face say this, computers didn't become unobtanium, they just saw a price increase of 20-30% for a full typical Gaming PC. in 2021 the prices were actually far higher, and a lot of Components, especially any remotely competent GPUs, were flat out UNAVAILABLE.
6. OP said that Humans were occasionally writing answers to prompts to improve accuracy, and said nothing of the training. Likewise I'd take that "third-world exploitation" with a huge grain of salt, it makes absolutely no sense whatsoever, as it's far easier to just make the whole World generate the training data for free, instead of limiting yourself to third world countries, and having to pay them for it.

[sockeye-d]

1. Even if I do, LLM-generated content is still inherently incomparably more plagiarized than human-created content.
2. Different AI, none of those developments are using large language models.
3. It is in fact not paywalled. You can minimize the banner and read the entire thing.
4. I actually do. Do you think GitHub is gonna be running their servers on a data center optimized for matrix multiplication? Likely no!
5. Both RAM and storage has increased in price, including spinning media. This DDR5 kit has increased in price $400 since 2025:
   
   I would call $100 a reasonable amount of money to spend on RAM, but I wouldn't call $500 a reasonable amount. A $1k (in 2025) PC build increasing in price 40% does actually put it out of reach of many people! And does it actually matter to the current hardware price inflation whether there was also a hardware shortage in 2021 as well?
6. That is certainly one interpretation of what they said. However, the workers do not create training data — they fine-tune the AI model's output in a process called reinforcement learning from human feedback.

[mrjustaguy]

1. Debatable. In fact I can think of ways to train an AI to code without ever seeing a single line of human made code, and I wouldn't be surprised a lot of the "good" coding agents out there use something like that. I do know the method has a name I just cannot think of it right now.
2. Yes but actually no. They're all using the same hardware (and math), it's only a question of what you decide to run on the hardware.
3. No I cannot.
4. No, you clearly do not. Yea Github ain't gonna run on those servers. You know what is going to run on those servers? Basically Any kind of Simulation (which is very relevant to science and engineering, medicine and so on) and there is (and always will be) ample demand for such heavy compute.
5. if you cannot shill out an additional 500 bux for a 4-5 year investment you my friend, have much, MUCH bigger things to worry about, and shouldn't be considering putting too much time into gaming, let alone investing heavily into it, and should instead figure out why the hell you are in such financial turmoil to begin with, and figure out what the hell to do to get out.
   Granted you may be simply priced out of what you're willing to pay, and not by actual financial difficulties. In that case, Tough luck, you'll just have to wait.
   The Hardware shortage of 21 matters in this context, as AI is being pinned as making things worse then back then when this is blatantly false.
   I'd also like to note that, once this demand dies down as the build outs slow down (and it will) you will start seeing some major Hardware improvements in the Consumer segment thanks to the enormous R&D being done specifically because of AI, and/or with AI.
6. This would have little impact on the case the OP stated.

[sockeye-d]

1. Even if the output is only transitively dependent on copyrighted works, it would still be dependent on those copyrighted works.
2. That's not what you said. You said "AI is itself a tool actively used to improve Desalination Tech". In this context, "AI" is being used to refer to large language models rather than the study of computer intelligence as a whole. Saying "the math is the same" misses a huge amount of difference in model architectures and training techniques, all of which have significant influence on what a model can do.
3. You literally can, but I have archived it here for your benefit:
   
4. All of those are designed without mass market appeal and need a tiny fraction of the total compute infrastructure being created.
5. Many people can't! I don't know why you're being so elitist about it, $500 is a lot extra to spend. $10 extra a month over five years can be the difference between living and starving.
6. I don't think it does, it highlights one of the many ethical shortcomings of major AI companies.

[mrjustaguy]

1. If this is the case, anybody that ever comes into any form of contact with copyrighted material should not be able to make anything in the field of the material in Question, which would be particularly problematic in the entertainment industry, as anybody that has seen Star Trek suddenly couldn't make any Sci-Fi set in space for example.
2. If it is being used in any way shape or form to the benefit of something, it is being actively used IMO, and that is the case, LLMs included as at this point they're likely being used as additional filters for the simulation data generated.
3. I didn't include where the Electricity comes from for a couple of reasons (which is where the pollution is coming here from from the skim read of it) and that is because there are many ways to get it and it is being used by basically every form of economic activity, and thus any form of increase in economic activity would produce the same problems, and at the end of the day this is a short term problem relatively speaking, which regardless of AI we need (and are) trying to deal with
4. In these markets you can NEVER have enough compute. You will ALWAYS find more to do with it to reap the benefits from it.
5. In a really low income country, sure. but I mean REALLY low income country. 5 years is 1825 days, so that's basically an additional 0.3€ a day that you're paying extra, which is such a meaninglessly low quantity compared to the typical day to day costs variances just from what you decide to eat a given day, and that's stuff people almost never actually think about unless they're actually counting every penny.
6. Are they actually ethical shortcomings tho? Say you are hiring a bunch of people in India to help train your AI. Yeah you're paying them little by the standards of your Home country, and it is hard work. But what are they paid relative to their own country? Are they not now getting food on the table thanks to that job? And are they not doing so in a relatively safe environment compared to say working in a Mine or sth?
   As far as I can tell, this is entirely a matter of perspective if it is good or not, and none of us here is qualified to say if it is good or bad, for we are not the people that are doing this work and we do not know how they actually feel about it or not, and what their lives would look without that job, and how they'd feel in that scenario compared to this one.

[Ivorforce]

LLMs plagiarize Code, potentially causing legal Troubles over Source Code Licensing.

Godot already prohibits non-menial use, which means that copyrightable code already cannot be LLM provided. Licensing is not a gap in our current AI policy.

LLM Compute Centers consume massive Amounts of Water, making many Organisms die from Dehydration.
LLM Compute Centers produce much toxic Waste Gas, making it hazardous to breathe for anyone living nearby.
LLM Vendors won’t stop buying out Computer Parts for as long as End Users use LLMs.
This even makes crappy Computers cost as much as a good Computer should cost.
This means i can’t make a Game with cool, even just moderately compute‑intensive Systems and Graphics, and expect potential Players to have a Computer that can play it.

LLM Outputs are not always Computer-generated.
Because LLM Vendors know their Computer Output is often inaccurate,
they also employ Slaves in third-World Countries to answer Queries accurately.
This is especially true for War‑related Queries in military AI.

These are all valid concerns, and I support anyone who takes action to fight for our societal rights in this matter.

However, neither is Godot's current policy the cause, nor will a change to it solve these problems.
Godot's current policy already prohibits most use of AI. Any further prohibition is difficult to enforce (at the cost of maintainer time and resources), highly divisive (compare bevy's situation), and is unlikely to have an impact on the problems you care about.

[Grisgram]

I love those Ai discussions.

This has nothing to do with Ai.
Is it better to copy/paste code from stack overflow? even many of the answers there, given by humans, have been coded by an Ai agent just because the stack overflow user is a rating addict and wants to answer as many questions as possible.

Fact is, not a single developer knows everything. Each and every one of us needs to "look up things" now and then. so why not challenge my idea against an Ai agent? Ask on stack overflow instead, so that somebody else answers the question with code from HIS Ai agent?

I think, there are two roads, you can walk on Ai code.
Road A: completely Ai-written PR's from contributors who don't have the required technical skills and who don't understand every piece of the code they submit. This road needs to be closed, I agree 100%

Road B: A contribution done by a human, and may have been bug-checked, code-reviewed by an Ai agent and it contains some lines or functions that the Ai could write with less memory footprint or performance overhead than the human. In short: some parts that have really been optimized by the Ai. From contributors who have the technical skill to judge, which implementation is the better one for the engine.
This road should be a Highway and open because it is:

• simple reality. more than 80% of developers cross-check their code with claude, gemini, gpt ,... you name it. almost nobody never touched an Ai for code
• Ai is here. And it will stay. You can't win this fight. 150 years ago "Cars" came. And stayed. You may like them or not, but they are here. And they do more harm and damage to the environment than Ai does. Forbid cars before you forbid Ai. Adapt to the new reality.

[mrjustaguy]

I'd like to point out that Cars kill a lot of people every year, at an estimated 1.1m deaths every year, and tens of millions get hurt (source - https://www.who.int/news-room/fact-sheets/detail/road-traffic-injuries)

That's a lot of Deaths due to Cars, yet we still use them all the time, and some even resist using them less when they don't Need to (ie don't really benefit from doing so in a given situation).

The Negative Side-effects of AI on human life, are comparatively small (though it Does pose a very real threat of hurting a LOT of people through criminal actions on Critical Infrastructure, but that will become less and less of a concern as time goes on and possible attack vectors get eliminated)

I would however also like to note that this Does have plenty to do with AI but for different reasons.

1. A lot of people that are actively pushing against AI tend to be people that are (at risk of) being displaced by it and having to go for new jobs (I myself am somewhat in this category, and I tell you, Spending a Decade of time preparing only to at the last seconds have the rug pulled is an AWFUL feeling)
2. Another group of people that aren't at any serious risk of being displaced by it are "feeling" for the others that are and are trying to fight for them (basically your typical activists, God I hate them so much, they just know how to make everything worse thinking they are helping)
3. Finally, there are people who are being directly impacted by Higher Costs of living, Noise etc. due to AI setting up camp near them

The main problem is just human nature. We really don't like feeling like we are getting screwed over (even if for a really short time), even if it is very likely to long term benefits us, and a lot of us are simply awful at adapting and/or letting go of what you need to and moving on.