There isn't really a reliable way to create a user fed "AI Tag". It is an impossible problem, stuff like this has been happening at all UGC companies.
Now, this is NOT a defending VRC post. That survey was one of the worst designed surveys I have seen in a long time, and you can tell this was made by someone who either isn't familiar on how to gather information or is incredibly behind on the state of the art. You cannot ask people if they want your company to add something that is impossible to achieve; and confuse those who do not know what goes behind all of the systems in place. They should have explained how it is impossible, instead of going "But you are not going to be able to see them :'C" so here it goes:
Slop is only identifiable when it has the "slop patterns"; and even then, a lot of the times it is impossible to differentiate or reliably enforce it. This is unless you want it to become panopticon and share your progress 24/7 everywhere (just for someone with AI copy it anyways). Right now, you can see the obvious signs, but what happens when models become increasingly good, to the point of indistinguishability? There are flows now where an AI generated model can be generated; go into a retopology engine; get auto UV unwrapped; and have their textures generated in a couple of minutes. Like just look at
devforum.roblox.com/t/textur… . Stuff like this becomes indiscernible from what someone who isn't the best at blender can do.
Not only that, but so many connectors and integrations have been made for it that watermarking has become, also, impossible. You can have an AI that generates text generate code to interface with blender, and have it create, work, and export models super easily. Because it is through code, you cannot reliably add an invisible watermark like SynthID into it; and even if you could, it would be really easy to reverse engineer and remove, just how we see it with SynthID.
I'll give ya an example, and that is the Human Verified program at Spotify. They have been trying to work an automatic way for identifying AI generated content for around 4 years now, under pressure of record labels (Allegedly). But the issue their engineers kept encountering is: how do you recognize and tag AGC? How do you make filters for it, stop "Voice Theft" WITHOUT hurting original artists?
AI songs have been and are flooding the platform, and record labels are not happy that they are taking a cut of their revenue (allegedly). Even stuff like Pangram, what is considered bleeding edge detection for AI prose, has been struggling with the latest of the latest, still can't reliably recognize code, and is in a constant back and forth battle to accurately detect stuff. So, detecting something is AI programmatically is next to impossible.
Hand tagging is also not the answer. the problem with it is, well, there is no incentive for creators to put it. Like for example, look at avatar tagging. I can tell you right now, most NSFW avatars out there do not carry the tag of NSFW out of fear of a ban. And yes, even if the tag ban is a myth, people still are afraid of it.
Creators are never going to be encouraged to use it, as why would they? Tagging will never work unless there is an incentive to the creator, or a punishment for the lie. And well punishment is impossible, while incentives for slop is disingenuous.
The only solution I see possible is an "algorithmic" solution. Get identifiers, of not only slop, but potential bad/good worlds in a per user basis. It would require VRC to gather more data yes (so it would need to be opt in), but being able to differentiate what you would like, dislike, etc. would do wonders for world discoverability, combined with a verified creator program. Something that looks further from just how long they have played the game, but also if they have a community, if they interact, etc.
Inadvertently, this could catch some of the slop identifiers in the process; if it is a well designed classifier and recommender. So say if a person usually enjoys and spends more time in worlds where FPS is kept high; topology count low; the world creator interacts with the community; and some other inferred world information they all could discern potential slop, even if its invisible to humans.
This has been the only path Spotify has been able to find around stuff like this, and EVEN THEN, it is not perfect. They are not able to catch all AI that comes in, but can help people filter it by trying to find hidden latents in the data and strive to recommend higher quality content.
artists.spotify.com/en/blog/… newsroom.spotify.com/2025-09… newsroom.spotify.com/2026-10…
I am pro-user choice. Just like in everything in life, users should be able to know what tools, tech, etc are used to build something (this is in all aspects of the tech, not only applies to AI). But, implementing anything like this technologically, this is quite literally a HERCULIAN task.