Roblox has decommissioned a researcher’s automated scraping agent after it collected content from 3D model-sharing sites, including adult material.
The company told CyberInsider that it never used the downloaded data to train AI models or for any other purpose.
The incident was disclosed on October 26 by Open3DLab administrator Salaryman, who reported that a crawler downloaded 1,027 files totaling 145.5GB from Open3DLab and sister sites SFMLab and SmutBase between October 15–16.
Roblox operates a gaming and creation platform where users build and share interactive experiences. The three affected sites are independently operated communities that offer free 3D resources, including characters, props, and environments, and creators choose licenses for their uploads. All three are intended for adults, while SmutBase exclusively hosts adult content.
In response to CyberInsider’s request for comment, Roblox confirmed that one of its researchers built the collection agent and said it malfunctioned.
“A Roblox researcher built an agent to collect publicly available 3D assets that mistakenly scraped content from Open3DLab, failing to operate as intended. The agent was decommissioned, and the data was not used for any purpose, including training any AI models. We are strengthening safeguards for how agents like this are built and reviewed, and will continue to do so as the agentic landscape evolves,” a Roblox spokesperson told CyberInsider.
Overwhelming download servers
According to Open3DLab’s report, the crawler accessed 18,926 project pages through 26,429 requests over 34 hours, effectively surveying almost the entire catalog.
It operated from a single Amazon Web Services address and identified itself using the user agent trellis2-data-sourcing/1.0, accompanied by a reference to “3D foundation model research” and a Roblox employee’s email address.
Of the 712 projects whose files were downloaded, 345 were classified as adult content, accounting for 62.6GB. That included 224 projects from SmutBase and another 121 marked as adult on SFMLab.
Salaryman said the crawler retrieved but ignored the sites’ robots.txt instructions, scraped despite terms prohibiting it, and collected files from 423 projects with noncommercial licenses.
The downloads also reportedly saturated the communities’ free download servers for a couple of days, disrupting access for other users.
Administrators discovered the activity on October 18 and blocked further requests that morning. The crawler returned on October 19 and 20, but its subsequent requests were blocked, according to the report. Open3DLab said Roblox did not respond to two emails seeking answers before publication.
Open3DLab suspected the collection supported Roblox’s AI research, citing the crawler’s identification and the company’s work on 3D foundation models.
Roblox’s Cube foundation model powers generative creation tools, including a “4D generation” beta introduced in February that adds functionality to generated 3D objects, such as making a car drivable.
Roblox’s statement directly denies any use of the scraped material, though there’s no explicit confirmation that the files were deleted.
Leave a Reply