Roblox Accused of Scraping 145GB of 3D Models for AI

Roblox is facing allegations that a web crawler linked to its AI research downloaded 145.5GB of 3D models from three community websites, including 62.6GB of adult content.

Open3DLab published the accusations on October 26, 2026, claiming that an automated crawler accessed Open3DLab, SFMLab, and SmutBase without permission.

According to the original Open3DLab investigation, the crawler examined 18,926 project pages through 26,429 requests over approximately 34 hours.

It allegedly downloaded 1,027 files from 712 projects, totaling 145.5GB. Of those projects, 345 contained adult content, accounting for 62.6GB of the collected files.

Open3DLab also reported that 423 downloaded projects carried licenses restricting commercial use.

The crawler identified itself as "trellis2-data-sourcing/1.0" and referenced research involving 3D foundation models. Its user-agent information included an email address associated with a Roblox researcher.

However, that identification does not independently establish who operated the crawler.

Open3DLab claims the automated system ignored its robots.txt instructions and terms of service. Administrators blocked further requests after detecting the activity on October 18, although additional attempts reportedly continued through October 20.

The website operators suspect the files were collected for AI training or evaluation, potentially connected to Roblox's ongoing work on generative 3D technology.

Roblox introduced its Cube AI research initiative in 2025, focusing on generating and understanding three-dimensional assets.

Nevertheless, Open3DLab acknowledges that it cannot prove the downloaded models were used to train an AI system.

The operators say they contacted Roblox twice without receiving a response. Roblox has not publicly confirmed responsibility for the reported activity.