3 websites tell that a Roblox crawler copied 145.5 GB of 3D files

Open3DLab, SFMLab and SmutBase are websites where creators share 3D models. They tell that a crawler (software that reads websites by itself) with a Roblox name copied 1,027 files from 712 projects in October 2026. The total is 145.5 GB. They tell that the crawler did not obey their robots.txt file, their terms of service or the licenses of the creators.

What happened

The crawler read the primary page of 18,926 projects. Then it copied full files. Of the 712 projects, 345 are adult content, and 423 have a license for non-commercial work only. The websites blocked it on 18 October. They tell that it started again on 19 and 20 October and tried to get more files.

The proof

The websites show their server logs. Each call had a user agent (the name of the software) with a note about 3D model research and a Roblox email address. All calls came from 1 Amazon Web Services address. The email address is of a researcher who is an author of a Roblox paper on 3D models.

What is not known

The websites cannot show the cause. They think that the crawler collected data to train a 3D model. They do not know if a model used a file. Roblox gave no reply to 2 emails.

Insight: Rules without a technical limit did not stop the crawler

The crawler read robots.txt, a file that tells software which parts it must not read, and did not obey it. The websites tell that the speed limits of their free download servers were the primary limit on the copy. It reads as a weak point: a rule does not stop software that collects data. Limits in the server do. Hosts of content must have those limits.