Roblox downloaded 145GB models from us (including adult content) without asking
Points and comments are a snapshot, not live.
Roblox's AI crawler downloaded 145 GB of 3D models from Open3DLab, including adult content, without permission.
On October 15-16, 2026, a Roblox-affiliated crawler (trellis2-data-sourcing/1.0) from a single AWS address downloaded 1,027 files (145.5 GB) from 712 projects across Open3DLab, SFMLab, and SmutBase. 345 of those projects were adult content. The crawler ignored robots.txt, Terms of Service, and creators' non-commercial licenses. Roblox did not respond to emails. The author speculates the data was for training Roblox's 3D foundation model (Cube). They highlight Roblox's history of child safety lawsuits and criticism for exploiting young developers.
The post argues AI companies commit copyright infringement at scale while individuals face ruinous penalties, calling for Roblox to delete the data and respect creator licenses.
What commenters are saying
Commenters broadly condemned Roblox's actions, with some focusing on technical mitigation: blocking datacenter IP ranges (AWS, Azure, GCP) to thwart scrapers, though noting residential proxies remain a challenge. Others questioned the adult-content emphasis, arguing that removing such data from training sets can reduce output quality, especially for human figures. A few dismissed the incident as routine scraping by a lone researcher, not a 'supercrawler.' Some criticized Roblox's child safety record and the broader AI industry's disregard for consent and copyright.