Roblox downloaded 145GB of assets including adult content for AI training
The world of 3D modeling, driven by platforms like Roblox, is increasingly merging with sophisticated generative artificial intelligence. But this convergence has also opened up complex, and sometimes thorny, ethical questions about data ownership and scraping practices.
An independent community focused on 3D modeling, Open3DLab, recently stepped in to address a major digital incident involving their platforms and the burgeoning AI landscape. They publicly disclosed that a web crawler, linked to the gaming giant Roblox, systematically mapped their sites and downloaded a staggering 145.5 gigabytes of data.
The operation targeted Open3DLab, along with sister sites SFMLab and SmutBase. The crawler pulled complete copies of 1,027 files from 712 projects, suggesting the data was aimed at mass collection for “AI training purposes.” The scope of the download was particularly notable, with the crawler securing 54.8 gigabytes from 224 projects hosted on SmutBase, and an additional 62.6 gigabytes from 345 projects marked by SFMLab.
How did this massive data sweep happen? The investigation revealed the crawler operated without respect for standard digital etiquette. It used a specific user agent, AWS address. Crucially, the crawler ignored the sites’ robots.txt files and explicitly disregarded creator licenses that prohibit non-commercial use.
Despite the explicit prohibitions against scraping, the crawler proceeded to read practically the entire catalogue of 18,926 projects in just 34 hours. Though the system was detected and blocked by the following morning, the operation persisted, with subsequent attempts made over the next two days to test rate limits.
The connection to Roblox’s AI ambitions adds another layer of complexity. Roblox is actively developing a core generative AI system for 3D modeling, including a 3D foundation model and beta launches for “4D generation” in 2026. The user agent used by the crawler contained an email address that matched a researcher listed on the Roblox Cube research paper, suggesting a direct link between the data collection and the AI development pipeline.
While Open3DLab cannot definitively prove the crawler’s intent, they note that the data was likely gathered for training or evaluation related to the 3D foundation model. This incident underscores the urgent need for clear guidelines regarding intellectual property and data handling in the rapidly evolving space where gaming, AI, and creative content intersect.
Credit
