TL;DR
Get privacy and security gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Open3DLab says a crawler using a Roblox-linked email address downloaded 1,027 files totaling 145.5GB from Open3DLab, SFMLab and SmutBase in October 2026. The site operators say the downloads included adult material and works licensed for non-commercial use; Roblox had not replied to their requests for comment, and the intended use remains unconfirmed.
Open3DLab says a crawler identifying itself as Roblox-affiliated downloaded 1,027 files totaling 145.5GB from Open3DLab, SFMLab and SmutBase in October, including adult models and files whose licenses restrict commercial use. The site operators say they blocked further requests and contacted Roblox, which had not responded by the time of their report; the purpose of the downloads has not been confirmed.
According to Open3DLab, the crawler visited the main pages of 18,926 projects across the three sites in 26,429 requests over 34 hours. It used the user-agent string “trellis2-data-sourcing/1.0” and included an email address at Roblox, the site operators said. Their logs showed traffic from a single Amazon Web Services address. Open3DLab reported that the crawler retrieved its sites’ robots.txt file but did not follow its instructions; the sites’ terms of service prohibit scraping.
The downloaded files came from 712 projects. Open3DLab said 364 files, totaling 54.8GB, came from 224 projects on SmutBase, which it describes as an adult-content-only site. Including adult-marked SFMLab projects, the operators classified 345 projects and 62.6GB of the downloaded material as adult content. They also said 423 projects in the sample had non-commercial-only licenses. These figures and license classifications are based on the site operators’ account; Roblox has not publicly confirmed the collection or explained whether the files were used.
Open3DLab said it detected the activity on October 18 and blocked requests that morning. It reported that the crawler resumed on October 19, with requests blocked, and returned again on October 20 before stopping at about 15:00 CEST. The operators said they emailed Roblox twice seeking comment and received no reply. They described the crawler’s activity and its apparent connection to Roblox based on the user-agent text and associated email, but did not publish an independent confirmation from Roblox.
Licenses and Adult Creator Work
The report raises questions about permission, licensing and data sourcing for 3D AI systems. The operators say many of the retrieved files were not licensed for commercial use, and that adult models were downloaded alongside other assets. If files were collected for training or evaluation, their inclusion could affect creators’ rights and expectations; however, the report does not establish that the assets entered a model or how they may have been handled after download.
The adult-content element also matters because the files came from sites and projects with explicit material, not solely from general-purpose 3D libraries. The available account does not show whether any material was processed, retained, shared or used in a product. Those distinctions are important: a logged download is evidence of access, not by itself proof of training, publication or commercial use.
Products Worth Considering
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
As an affiliate, we earn on qualifying purchases.
Roblox’s 3D Model Research
Open3DLab links the incident to Roblox’s work on Cube, its generative system for 3D and 4D content. Roblox introduced Cube in March 2025 and described it as a core generative AI system. The paper accompanying its announcement discussed a 3D foundation model and future capabilities including character rigging and character or avatar generation. According to that paper, an early shape model was trained on about 1.5 million assets from licensed and public datasets, as well as free Creator Store assets from creators who opted in.
Roblox began rolling out 4D generation in February 2026, according to Open3DLab’s account. The site operators say the email embedded in the crawler’s user agent belongs to a researcher listed among the authors of the Cube paper and whose professional profile describes work on a 3D foundation model. That association is part of their reasoning, not confirmation from Roblox that the researcher or Cube was involved in this collection.
Open3DLab also says “trellis2” in the crawler’s label might refer to TRELLIS.2, an open 3D generation model released by Microsoft in late 2025. It acknowledges that the name could instead be an internal project label. The operators say the public TRELLIS.2 data toolkit does not include a web crawler or that user-agent string, leaving the relationship uncertain.
““145.5 GB of our creators’ work””
— Open3DLab
Purpose and Use Remain Unconfirmed
Roblox has not publicly explained the crawler’s activity in the material provided, and Open3DLab says its two requests for comment received no reply. The crawler’s user-agent text and email address suggest a Roblox connection, but the report does not include a company statement confirming who operated it or authorizing its work.
It is also unknown whether the files were collected for training, evaluation or another purpose; whether they were retained after the sites blocked access; and whether any were incorporated into a model. Open3DLab says training or evaluation is its most likely explanation, based on Roblox’s 3D model research and the nature of the files, while stressing that it cannot prove that account. The report says a fuller review of the site logs would follow.
Further Logs and Roblox Response
Open3DLab said it planned to publish a more detailed account of its logs at a later date. That could clarify which projects and files were accessed, the timing of requests, and how the sites distinguished completed downloads from blocked attempts. The site operators say their blocks stopped the crawler’s later requests, but have not provided evidence in the supplied report about what happened to files already retrieved.
The key outstanding development is a response from Roblox addressing whether the crawler was operated or authorized by the company, what the collection was for, and how the downloaded material was handled. Until the company or further evidence clarifies those points, claims about AI training remain an attributed explanation rather than a confirmed use.
Key Questions
What does Open3DLab say the crawler downloaded?
Open3DLab reports that it downloaded 1,027 files totaling 145.5GB from 712 projects across Open3DLab, SFMLab and SmutBase.
Did the downloads include adult content?
According to Open3DLab, 364 files totaling 54.8GB came from SmutBase, which hosts adult content, and adult-marked SFMLab projects brought the total classified as adult to 345 projects and 62.6GB.
Has Roblox confirmed that it collected the files for AI training?
No. Open3DLab says it suspects the collection was for training or evaluation based on the crawler label and Roblox’s 3D model research, but says it cannot prove that purpose. The supplied report says Roblox did not respond to the site’s requests for comment.
Were the files used to train a Roblox model?
That is not established by the report. Open3DLab says it does not know whether any downloaded files ended up in a model, and provides no confirmation that the files were used for training, evaluation or a product.
Source: open3dlab.com
