An independent 3D modeling community has accused Roblox of quietly harvesting its entire catalog. Open3DLab says a crawler linked to the gaming giant downloaded 145.5GB of 3D assets from its site and its sister platforms, SFMLab and SmutBase, in a scraping campaign that ran for days in October.

What You Need to Know

Open3DLab is a community platform where independent artists publish 3D models under licenses that often bar commercial use. Roblox has been building a 3D foundation model called Roblox Cube since March 2025, and the crawler's user agent explicitly cited "3D foundation model research." The dispute highlights a growing tension between AI developers seeking training data and the creators whose work is being swept up without consent.

The Scraping Campaign and Its Scale

According to a blog post published by Open3DLab, staff noticed a crawler "making a large number of requests to our sites and download infrastructure" on October 18. The crawler identified itself with the user agent trellis2-data-sourcing/1.0 (3D foundation model research; [redacted]@roblox.com) and originated from a single Amazon AWS address.

The operation was extensive. The crawler read "practically [its] entire catalogue" of 18,926 projects over 34 hours across 26,429 requests. It then downloaded complete copies of 1,027 files from 712 projects before suspending operations.

145.5GB

Total data downloaded

1,027

Files copied in full

18,926

Projects scanned

The crawler did not respect the site's robots.txt file or Terms of Service, which explicitly forbid scraping. It also ignored creator licenses for the projects, which prohibit non-commercial use. The downloaded files included 364 files, or 54.8GB worth, from 224 projects hosted on SmutBase, Open3DLab's sister site focused on adult content. A further 345 adult-marked SFMLab projects, or 62.6GB worth, were also taken.

Detection, Blocking and Persistent Retries

Open3DLab detected the scraping on the same day and blocked it by the following morning. Further download attempts were detected on October 19, "neatly within office hours, only briefly pausing for an hour around lunchtime," and despite being blocked, "did not reduce its download attempts."

A final attempt on October 20, with requests tapering, was "likely in an effort to test... rate limiting," according to the report. Requests stopped completely later that day. Open3DLab made another request for comment at that point, although it received no reply.

The community has since made changes to try to avoid a repeat. The crawler's "main limiting factor" was its use of free download servers that capped speed and concurrent downloads, Open3DLab noted.

Roblox Cube and the 3D Foundation Model Race

The user agent's mention of "3D foundation model research" points directly to Roblox's own AI ambitions. Roblox has been building such a model since March 2025, when it introduced Roblox Cube, its "core generative AI system" for 3D modeling and character rigging, with characters and avatars listed as future goals.

Further AI technology that could build in-game models using natural language prompts, called "4D generation," was beta launched 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 redacted the name.

Noting Roblox's "track record," Open3DLab's "most likely explanation" is that the crawler was collecting "training or evaluation data" for its 3D foundation model. The community's suspicion is grounded in a broader pattern: AI developers have increasingly turned to scraping public repositories, often without explicit permission, to feed data-hungry models.

Why This Matters

The dispute exposes a widening fault line between AI developers and the creative communities whose work fuels their models. For independent artists on Open3DLab, SFMLab and SmutBase, the incident is a direct threat to their livelihoods and licensing terms. If a major platform can ignore robots.txt and creator licenses without consequence, the entire system of voluntary consent that underpins public repositories is undermined.

For Roblox, the reputational risk is significant. The company is positioning Roblox Cube as a cornerstone of its creator economy, yet the crawler's behavior suggests a willingness to bypass community norms. How Roblox responds, whether with transparency, licensing deals or silence, will shape how creators view its AI ambitions.

More broadly, the case could accelerate calls for clearer legal frameworks around AI training data. As 3D asset repositories become valuable training grounds for spatial intelligence, the question of who owns and controls that data will only grow more urgent.