On October 15-16, 2026, a crawler identifying as Roblox downloaded 1,027 files from 712 projects across Open3DLab, SFMLab and Smut Base, totaling 145.5 GB for AI training purposes. The crawler used user agent 'trellis2-data-sourcing/1.0' from an AWS address, reading 18,926 project pages in 34 hours. It violated robots.txt, Terms of Service, and creator licenses, including 423 non-commercial projects.
Of the 712 projects, 345 (62.6 GB) contained adult content, including 364 files (54.8 GB) from Smut Base. The crawler respected no speed limits and resumed activity after being blocked on October 18. Roblox confirmed the email belongs to a researcher on the Cube paper team, their core generative AI system for 3D/4D.
The sites blocked the crawler same morning, but it returned October 19th, testing rate limits before ending October 20th. No response was received from Roblox. The incident raises concerns about AI training data acquisition and creator rights.
Source: Hacker News · Summarized by HeadlinesBriefing