On this point, we still have to return to the question of how to identify it. No matter how bad AI is, no matter how bad it gets, if effective identification cannot be achieved, then all the so-called “clearer policies” are empty talk. What exactly do you expect VRChat to do? Issue a statement saying, “We do not welcome artificial intelligence. Although we cannot effectively identify it, we still ask those who use AI to voluntarily get the hell out”? I have listened to you go on for quite a while about the legitimate reasons for opposing AI, and I agree with most of your views with unparalleled conviction. But apart from proposing “AI labels” and “filters”—these utterly meaningless and comical suggestions—do you have any other methods?
I’d like to try and articulate my stance on AI, and I think some people in here might find this helpful for one reason or another:
First off, “AI” as its known currently is not artificial intelligence. Its snake oil being presented as AI. Its an empathy/dopamine simuator with access all of the internet. If you think its sentient, I feel sorry for you, truly.
LLMs are genuinely cool technology, but the way they are being developed is abhorrent, with absolute disregard for environmental, social and psychological impacts, disregard for peoples job security and art, without consent of artists or anyone they’ve scraped data from. The technology at its core spits in the face of art and artists. That includes programmers.
Furthermore, people who are lucky enough to retain jobs in the tech industry are being forced to use this technology against their will. Their only option is to use it, or lose their job in a job market where tens of thousands have already been laid off. This technology is being forced into the tech industry, and from there every facet of life. Everywhere it doesn’t belong, because the people who are developing it want it to become a dependency of normal life.
Workers in the industry cannot push back against the encroachment of AI, and its all but consumed them already. WE CAN push back against the encroachment of AI in creative spaces, and that is why is so important that we do. Other places definitely have avenues for pushback against AI and data centers and such, pushes for regulation etc. but you can’t fight every battle, and pushing back in creative spaces as just as important.
This is a moral and ethical disagreement with AI and the way it is wrecking society right now.
AI could be a cool tool for development, but until the problems these billionaires are creating are remedied, it must be rejected. This is my personal line.
Right now, prices are so unsustainably low, and the tech is so available and intentionally THIRSTY for more users, it is leading to a deluge low effort AI slop, and low effort plagiarism that is encouraged by the system. More users, at any cost.
They are also openly pushing for US gov intervention, anti trust law immunity, regulation immunity, bans on open source models, and other lawful protections that are meant to protect their financial interests. Everything surrounding this is corrupt, and to see that corruption flooding into VRChat is just so tragic.
Generative AI, as it currently exists, is theft. Full stop. Artists have had their styles completely ripped off, and in some cases lost their entire livelihoods because all the AI slop looks like their art. Tons of people are taking art work commissions and not disclosing that they are generating the art, or generating + tracing over, and these people are flooding artistic spaces and pushing out real artists.
Indie developers are having their entire games stolen and ransomed. The video game decomp scene completely shut down because it got flooded with poor quality AI decomps. AI is ruining every space it touches. VRChat is not immune to this.
1 Like
I completely agree with what you said. But you should be protesting to the AI training companies instead of hounding a game company relentlessly, pressuring them into introducing so-called “relevant restriction policies” for which you yourselves can’t even come up with a good plan.
1 Like
My suggested solution to the AI labeling problem:
Split it in 3 labels:
- Fully generated
- AI tool assisted
- Minor AI usage
and include an optional field for uploaders to write how they used AI
allow all 3 of these to be filtered individually, let the users choose their own line.
Also VRChat really needs to understand, THIS IS THE COMPROMISE POSITION.
The no compromise position is a full ban on AI content, and if you keep denying the compromise, you’re going to either get people pushing for the full ban, or saying goodbye to your platform in short order.
it seems you missed this paragraph
Workers in the industry cannot push back against the encroachment of AI, and its all but consumed them already. WE CAN push back against the encroachment of AI in creative spaces, and that is why is so important that we do. Other places definitely have avenues for pushback against AI and data centers and such, pushes for regulation etc. but you can’t fight every battle, and pushing back in creative spaces as just as important.
I think the VRChat team has already made their reasons for ultimately deciding not to do this quite clear. Without an effective identification method, such a label could only rely entirely on the author’s self-disclosure. But once that door is opened, there will inevitably be creators who get harassed and attacked simply because they are suspected of using AI. Your proposal would not ease the problem of AI-generated content; instead, it would create even more conflict. In fact, there is already a very clear example of this. Not long ago, a VRChat-like game called Resonite had this kind of nasty incident happen, and you can go look into it.
What if VRChat introduced an AI provenance detection system directly into its Unity SDK or Creator Companion? Rather than relying entirely on external AI detection tools, the system could inspect uploaded assets during the world-building and upload process.
For example, it could check for AI-generation metadata, embedded watermarks, digital provenance records, and other identifiable information associated with supported AI generation tools. Creators could also declare AI involvement in their projects, with the system collecting this information into an asset manifest during the build process. Ideally, this information would be preserved and verified when the world is uploaded to VRChat.
Of course, this wouldn’t be foolproof. Metadata can be removed, watermarks can be manipulated, and not every AI generator provides identifiable provenance information. That’s why I wouldn’t suggest relying on a single detection method or automatically punishing creators based on an AI detector’s results. However, combining provenance verification, creator declarations, and automated asset inspection could provide a considerably stronger system than simply asking users to voluntarily apply an AI label.
I think the important distinction here is that we’re not necessarily trying to create a system that magically identifies every AI-generated asset. We’re trying to make AI involvement more traceable, make deliberate misrepresentation more difficult, and give VRChat a reasonable foundation for reviewing questionable uploads
1 Like
Resonite has a robust AI policy in the works, give it a read.
Re
1 Like
If you know enough about AI generation, you’ll realize that these detection methods are not very effective, and if they were actually to work, they would instead compromise creators’ information security.
That would also imply that the vrchat Devs have low knowledge of coding . I’m not saying it has to be perfect because nothing is perfect .
Thanks for sharing. I’ve taken a brief look at the draft, and I respect the Resonite team’s willingness to tackle this issue. The draft does address evidence requirements, false reports, and harassment, and it includes some safeguards.
However, how well those measures will work in practice remains to be seen. How much benefit can a disclosure system with limited coverage actually provide? How many resources will verification and dispute resolution require? And how effectively will it reduce false accusations and harassment in practice? These are questions that still need to be evaluated. Writing safeguards into a policy is an important step; consistently enforcing them and achieving the intended results will take practical experience.
So, while I appreciate Resonite’s willingness to explore this approach, I still think VRChat’s concerns about enforcement costs and community risks are entirely reasonable. Keep in mind that VRChat is already a relatively mature game, with an active user base on a completely different scale from Resonite’s. Its user demographics and community makeup are also vastly different from Resonite’s almost entirely creator-based community—and even there, AI restrictions have already sparked ugly incidents.
Resonite’s draft offers an approach worth watching. Although I’m pessimistic, I would be glad to see them find a workable path forward. At this stage, though, the existence of that draft is not enough to demonstrate that VRChat made the wrong decision.
It’s not a question of whether it’s perfect or not—it’s that the solution you proposed is simply unrealistic.
Two problems: detecting AI via automated means is getting harder and harder (even if you can kinda just look at it and tell), and- as EAC has shown us- putting the means of detection into the hands of the user, who has direct hardware access as well as unlimited access to the filesystem of the device these checks are done on, is a really futile cat-and-mouse game from a technical standpoint. If this would be a good solution, we wouldn’t currently have an avatar stealer bot problem. Evading detection is something AI companies and cheat makers have in common.
Yes.Unless AI training companies can be successfully pressured into ensuring that all AI-generated content carries a distinctive marker, you will forever be limited to trying to detect AI generation by its stylistic ‘flavor’ or by certain AI-specific quirks that will soon be patched through iteration. Even then, only a small subset of assets can be identified as AI-generated with any relative reliability
Well maybe they should try something different than EAC as it shows that the program doesn’t work . There so many new avatar ripping tools that have appeared since then and EAC has done nothing about it , let alone the vrchat Dev team
To be honest, if someone could develop a sufficiently effective automated detection filter and keep it effective over the long term, they should go work at OpenAI or Anthropic rather than stay at VRChat earning a pitiful salary.
OpenAI and Anthropic have incentive for something like that not to exist and not work, given they already try to make it harder to detect their output (and scraper bots- I see you pretending to be cURL, Claude, you slimy little snake >:U). Their alleged concern for their impact of their products on people is performative at best as well.
Ai badge is a red herring though, what people dont realize and what they really should be asking for is some sort of regulatory aspect of worlds. Currently there is none, and searching for worlds on VRChat is still a nightmare.
Have any of you tried searching for a world using vrchat native menu? it’s aweful, and the search function features OR logic instead of AND. If you search for a “Cat” and “Dog” in the world menu, vrchat search results with all the worlds that have “Cat” anywhere in the name, description, tags, and all the worlds that have “dog”. It is not possible to narrow down results with more tags. No, it only shows more results the more tags you input. That is absurd.
And this doesnt’ combat the countless of UnityArchViz.Prefab worlds uploaded every where. There is nothing NOTHING to let users know the world they are about to enter is worth visiting or not. and all that does is waste everyone’s time and it’s rarely rewarding. This has taught the larger part of the users on vrchat that world exploration is not worth the time investment.
All the serious creators have mostly left the platfrom at this point. And why would anyone want to create for vrchat at this point? The black Cat, Murder 4, and popcorn palace have been at the top and recieve most of the traffic of online users for the past 6 years. It’s never going to change.
2 Likes
Right now, everyone’s just beating around the bush that VRChat just ignored 95% of the playerbase.
A real solution would be an investor putting in a nicely-worded email to put a stop to this, or Steam actually enforcing their slop rule.
But neither of those are going to happen. So, chill.
With respect “just accept it” feels spineless. It’s a damaging technology that is literally destroying rare books and jacking up the price of everything. We have a moral obligation to tell any and every vapid uncreative using AI that they are making the world worse.
3 Likes