Roblox downloaded 145GB models from us (including adult content) without asking

choult 464 points 268 comments October 27, 2026
open3dlab.com · View on Hacker News

Discussion Highlights (20 comments)

hypfer

Only tangentially related, but I've recently realized that unless you're some sort of business with a valid requirement of that, you can just block incoming traffic from all of AWS, DigitalOcean, Azure, etc. I think the spark came from some other HN comment I saw? So thank you for that, whoever wrote that. That alone is not a sufficient strategy of course, but you can get rid of a lot of bots by just refusing to talk to where they tend to live. Of course, if everyone were to do that, they'd eventually pivot to residential proxies. But I don't see that happening.

GL26

maybe stupid question but how is it possible to not be protected against supercrawlers like this, when you own large amounts of data, specially if all requests are coming from the same server ? Aren't cyberprotection services like CloudFlare work against this kind of "attacks" ?

caaqil

Who actually cares? Isn't the whole premise of the AI revolution just gorging ourselves with other people's data so we can have AGI? I was under the impression that we have accepted this sacrifice with OAI and friends leading the way.

KolmogorovComp

I'm sympatic to open3dlab but I think the focus on adult-content is just trying to spark outcry for cheap. It has been shown times and times against that removing adult content from the training source (at least for diffusion models), results in generally lower-quality output, even outside of adult-content, particularly in terms of representing human bodies.

swiftcoder

I wonder at what point the Roblox kids all reach 18 years old, and Roblox can pivot them all to a separate adult version thereof...

peridotintellig

I'm not even impressed..

bijowo1676

if you put stuff available to download, dont be surprised that someone will download it What is the actual news here, can anyone explain ????

j-pb

3$ in outbound traffic on a public service, how is this news?

semiquaver

So what? You made your content available and someone downloaded it. Why is this even notable?

bsenftner

If that data is at all important, why is is downloadable?

speedgoose

145GB isn’t much. It fits in RAM in some computers.

echoangle

Couldn't the adult file download just as well be used to train a model that rejects adult uploads to roblox? I don't see the outrage.

littlecranky67

I know it is an unpopular opinion, but if you do not want your content to be scraped, you have to put it behind payment walls and copyright it. In Germany we have decisions from the highest courts that deemed websracping perfectly legal (it was in the context of a competitor scraping airline ticket fares).

philipwhiuk

I assume they need the adult content in order to detect adult content generated by their users.

origini

I'm just to ignorant to this in context of how the reality just works. Downloading content on the web is what the web is for. Using the data for training is something lawers need to figure out. open3dlab should spend the time and energy, as unfortunate it is, to sue them to get this resolved. Ablog post about this is basically 0 information. The stories about soooo many bots now crawling the internet and creating a lot more traffic is going through news and hn for month.

phn

How can you start an outcry like this, but host stuff like this, tagged with a permissive license: https://open3dlab.com/project/e2a0830d-f166-4b1b-aabf-8ca735... Just an example, there's plenty more on the frontpage. I have nothing against these guys, but I just can't understand the moral high ground of the complaint.

andreygrehov

The internet must understand one simple rule. What’s downloadable will always be downloaded. There are plenty of people who don’t give a flying fk about your LICENSE.md, robots.txt, ToS and whatnot. It’s like leaving your bike unlocked in the public. Even if there is a giant cardboard next to it saying: “please, don’t steal the bike” - nobody cares. Someone will strikethrough the “don’t” and grab the bike.

arshxyz

Scientists say this is the first occurance of the word "model" in the title of a post on HN that does not refer to a neural network of some sort since 2023

lifeisloving

Everyone should be hosting a poison fountain and pressing thumbs down on models when they provide really good answers.

paxy

Scraping is legal, last I checked. If you don't want people downloading your content then don't host it.

Semantic search powered by Rivestack pgvector
10,780 stories · 101,661 chunks indexed