Open3DLab accuses Roblox of downloading 145GB of 3D assets for AI training purposes – automated crawler ignored site rules to grab over 1,000 files, including adult content, during scraping spree
Follow us
Add us as a preferred source on Google
Subscribe to our newsletter
Independent 3D modeling community Open3DLab has publicly warned that a web crawler linked to the gaming platform Roblox has mapped “almost every project” on its website, along with sister sites SFMLab and SmutBase, downloading 145.5GB of data in the process. In a blog post, Open3DLab reports that “complete copies” of 1,027 files from 712 projects were found in the process, which it believes were taken for “AI training purposes.”
Go deeper with TH Premium: AI in data centers
(Image credit: Microsoft)
The blog post goes into the specifics of the incident, along with some changes Open3DLab has since made to try to prevent a repeat. According to the post, Open3DLab staff noticed a crawler “downloading a large number of requests to our sites and infrastructure” on October 18. The crawler used trellis2-data-sourcing/1.0 (3D foundation model research; [redacted]@roblox.com) as a user agent, with the data coming from a single Amazon AWS address. The crawler did not respect the site’s robots.txt file or the terms of service, which explicitly prohibit crawling. It also, according to the report, ignored the creator’s licenses for the projects, which prohibit non-commercial use.
The crawler reads “handy [its] entire catalog” of 18,926 projects in a span of 34 hours over a total of 26,429 requests. Once complete, the crawler downloaded 1,027 files from 712 projects before “suspending operations.” Files downloaded included 364 files, or 54.8GB worth, from 224 open SmutBs projects hosted on SmutB Projects adult content A further 345 projects of adult marked SFMLab projects, or 62.6GB worth, were also taken.
Latest Videos FromTom’s hardware
According to Open3DLab, the crawler’s “main limiting factor” was its use of free download servers that limited speed and concurrent downloads. The hack was detected the same day, and blocked by the next morning. Further download attempts were detected on October 19, “all during office hours, only a short break for an hour around lunchtime”, and despite the block, “his download attempts did not decrease.”
A final attempt on October 20, with requests tapering, “probably in an effort to test … rate limiting,” was made, with requests stopped later that day. Open3DLab made another request for comment at this point, though without a response.
The Roblox user agent mentions “3D foundation model research.” As Open3DLab’s blog post notes, Roblox has been building such a model since March 2025, when it introduced the Roblox Cube, its “core generative AI system” for 3D models for character rigging, with characters and avatars listed as future goals. Further AI technology that can build in-game models with natural language prompts, called “4D Generation”, was launched in beta in February 2026. The email address listed in the user agent matches one of the researchers listed on the Roblox Cube research paper, according to Open3DLab, although it has redacted the name here.
Noting Roblox’s “track record,” Open3DLab’s “likely explanation” is that the crawler collected “training or evaluation data” for its 3D foundation model, though it notes it can’t prove it.
Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.
Roblox is a gaming platform and game creation system with over 33 million users and billions of “game experiences” for its player base, 40% of which are under 12 years old. Open3DLab is clear to say that it “does not believe that Roblox is committed to collecting adult content”, but that “as far as [its] Logs show, no one checked,” which, it believes, is “not a good position” for Roblox to be in.
follow Tom’s Hardware on Google Newsor add us as a preferred sourceto get our latest news, analysis, & reviews in your feeds.
