A web crawler identifying itself as a Roblox research tool allegedly copied 145.5GB of 3D models from three community sites, including large amounts of adult material, in October 2026, according to a technical report published by Open3DLab on 26 October.
The site's operators say server logs show thousands of automated requests from the crawler to Open3DLab, SFMLab and SmutBase, resulting in more than a thousand file downloads from hundreds of projects over several days.
They suspect the activity was linked to AI research, but have not established who ran the crawler or how the downloaded files were used.
Open3DLab says its logs recorded activity between 15 and 20 October, including downloads from hundreds of projects.
Although the crawler's identification string contained a Roblox email address, the operators accept that this does not in itself prove the company was responsible.
Crawler Allegedly Lifted 145.5GB of Community Models
Open3DLab says it detected the crawler on 18 October after noticing extensive automated activity across the three websites and their download infrastructure. The tool used the identification string trellis2-data-sourcing/1.0, describing itself as being associated with 3D foundation model research.
The site's operators report that the crawler accessed 18,926 project pages through 26,429 requests over 34 hours. It subsequently downloaded 1,027 files from 712 projects, amounting to 145.5GB.
The downloaded material reportedly included 364 files totalling 54.8GB from 224 SmutBase projects. SmutBase hosts adult content exclusively, while Open3DLab's analysis also identified adult-labelled projects on SFMLab.
Roblox downloaded 145GB models from us (including adult content) without asking https://t.co/c0c6F70Ud8 (https://t.co/hwoqJGHFZw)
— Hacker News 50 (@betterhn50) October 27, 2026
Open3DLab classified 345 downloaded projects as adult content, totalling 62.6GB, although the figures have not been independently verified.
The site also alleges the crawler ignored its anti-scraping rules. Of the 712 projects downloaded, 423 reportedly carried non-commercial licences, raising questions about reuse without automatically establishing copyright infringement.
Roblox Link Remains Unproven
Open3DLab suspects the activity may have been connected to Roblox's generative AI research, particularly its Cube system, which the company introduced in March 2025 for generating three-dimensional and four-dimensional content.
The operators point to the crawler's identification string and an email address they say belongs to a researcher associated with a published Cube research paper.
They also suggest that the 'trellis2' portion of the identifier could refer to Microsoft's TRELLIS.2 3D-generation model.
These details do not establish who operated the crawler or why. Open3DLab says the requests came from an Amazon Web Services address, while the user-agent string's Roblox identification remains unverified. The site suspects AI data collection but cannot confirm whether the files were used to train or evaluate a model or influenced any finished product.
Adult Content Makes up 62.6GB of Downloads
The reported volume of adult material raises a separate question about the selection of files for potential AI research. Open3DLab says the crawler downloaded projects from SmutBase and adult-labelled SFMLab pages without any apparent distinction between them and other assets.
The operators have not established that the crawler deliberately targeted explicit material. Nor does their report demonstrate that the files entered a Roblox product or that children encountered content generated from them.
The distinction matters because downloading a file and incorporating it into a model are different stages of a process. The available account does not establish that the latter occurred.
Open3DLab warns that downloaded assets could persist even after creators delete their original projects. One project has since been disabled on SFMLab, but the site cannot confirm whether copies remain elsewhere.
Site Presses Roblox for an Explanation
Open3DLab says it first contacted Roblox on 18 October and sent a follow-up after the crawler resumed activity on 19 October. The site's operators report that subsequent requests were blocked and that the final recorded activity ended at around 3pm on 20 October.
They want clarification on whether Roblox operated the crawler, what the downloaded files were intended for and whether any were used to train or evaluate an AI model. They are also seeking answers about the handling of adult material, the retention or deletion of downloaded files and compliance with creators' non-commercial licences.
As of 26 October, Open3DLab said it had received no response to its requests for comment. The operators say they have retained the relevant server logs for potential independent auditing.
The report provides figures and technical details about the activity observed by the sites, but the crawler's operator, its intended purpose and the subsequent use of the files remain unconfirmed.
