Open3DLab has accused Roblox of scraping 145GB of 3D assets from its platform to train AI models. According to the company, an automated crawler processed nearly its entire catalog of 18,926 projects over 34 hours, executing 26,429 requests. The operation resulted in the download of 1,027 files from 712 projects, including significant amounts of adult content from affiliated sites.
The scraping activity was detected and blocked by Open3DLab, though subsequent attempts to access the data continued through October 20. Open3DLab identified the crawler's user agent as referencing 3D foundation model research and noted that the associated email address matches a researcher linked to the Roblox Cube project. Roblox has been developing generative AI systems for 3D modeling since March 2025.
While Open3DLab suggests the data was likely collected for training or evaluation purposes for Roblox's AI initiatives, it acknowledges it cannot definitively prove the intent. The platform emphasized that the automated process failed to filter out adult content, which constitutes a notable portion of the retrieved data. Roblox has not provided a comment on the allegations.
