Open3DLab says Roblox crawler downloaded 145.5 GB of models without permission

Open3DLab, SFMLab and SmutBase say a crawler identifying itself as belonging to Roblox downloaded 145.5 GB of 3D models from their sites on October 15 and 16, 2026. The operators say the crawler read 18,926 project pages and downloaded 1,027 files from 712 projects, including 345 adult projects, while ignoring robots.txt, site terms and creators' licenses. They blocked further attempts and say Roblox has not replied to their emails.

What happened and how they responded

The crawler used the user agent trellis2-data-sourcing/1.0 (3D foundation model research; [redacted]@roblox.com) from a single Amazon Web Services address. It read 18,926 project pages in 26,429 requests over 34 hours, then downloaded 1,027 files from 712 projects, including 364 files (54.8 GB) from 224 SmutBase projects; with SFMLab adult projects, 345 of 712 projects (62.6 GB) were adult content. It retrieved robots.txt but did not respect it, ignored Terms of Service that forbid scraping, and 423 projects were licensed non-commercially. The operators say they found it on October 18, blocked further attempts, and emailed Roblox without reply. The crawler resumed on October 19 and 20 with all requests blocked, with activity on October 20 from 8:00 to about 15:00.

Why they think it may be for AI training

Roblox has been building a 3D foundation model since at least 2025. It introduced Cube in March 2025, with a paper describing a foundation model for 3D intelligence; its first shape model was trained on roughly 1.5 million 3D assets. In February 2026 it began rolling out 4D generation built on Cube. The user-agent address belongs to a researcher among the 40-plus authors of the Cube paper. The 'trellis2' label may refer to TRELLIS.2, an open 3D generation model Microsoft released in late 2025. The operators say the most likely explanation is collection of training or evaluation data for Roblox's 3D foundation model work, but they cannot prove it and do not know whether any files reached a model.

Adult content, broader context and questions

The operators say more than a third of Roblox's daily users were under 13 in 2025 by its own figures and that its rules ban sexual content. They say the crawler took 224 SmutBase projects and 121 more marked adult on SFMLab, and that 345 of 712 downloaded projects were adult content. They state they do not think Roblox set out to collect adult content and have no evidence it reached a model or product, but say nobody checked. They object to adult sites being emptied into training data without permission and mixed into content offered to children. They also note Meta's AI crawler made about 380,000 requests and loaded 211,000 project pages before being blocked. They ask Roblox whether it ran the crawler, what the data was for, whether it was used to train or evaluate a model, why it ignored rules, whether sites were reviewed, whether adult content was excluded, whether files and derivatives will be deleted, and whether license choices will be respected. Open3DLab, SFMLab and SmutBase are three community sites for free 3D resources run by a small team; all are intended for adults only.

Sources: open3dlab.com