Open3DLab has accused a web crawler identifying itself as connected to Roblox of downloading approximately 145.5GB of community-created 3D assets without permission. The site’s operator alleges that the crawler collected 1,027 files from 712 projects across three 3D modeling websites, potentially for use in generative AI research or training.
The allegations were published on October 26, 2026, by Salaryman, the operator of Open3DLab, SFMLab, and SmutBase. The operator says server logs document extensive automated scraping activity between October 15 and 20, including downloads that allegedly disregarded website restrictions and creators’ licensing terms.
However, Roblox has not publicly confirmed operating the crawler, and there is no independent confirmation that the downloaded assets were used to train an AI model.
According to Open3DLab, the crawler initially accessed 18,926 project pages through 26,429 requests over approximately 34 hours. It subsequently downloaded complete copies of 1,027 files associated with 712 projects, totaling 145.5GB.
The automated requests used a user-agent identifier beginning with “trellis2-data-sourcing/1.0” and describing its activity as research involving 3D foundation models. The identifier also contained an email address associated with the roblox.com domain.
Open3DLab says the crawler operated through a single Amazon Web Services IP address rather than infrastructure directly associated with Roblox. The operator also claims that the email address belonged to a researcher involved in Roblox’s published 3D foundation model research.
Those details suggest a possible connection to Roblox, but they do not conclusively establish who operated the crawler. User-agent identifiers can be configured to contain arbitrary information, and Open3DLab acknowledges that its attribution remains unconfirmed.
The operator alleges that the crawler retrieved the websites’ robots.txt instructions but continued accessing material despite restrictions. Open3DLab also says the activity violated its terms of service, which prohibit scraping.
Of the 712 downloaded projects, 423 reportedly carried licenses restricting their use to non-commercial purposes. Because the sites host assets uploaded by independent creators, these restrictions raise questions about whether the downloaded material could legally be incorporated into commercial AI development.
However, downloading an asset does not by itself establish that it was subsequently used commercially or that copyright infringement occurred.
Another concern involves the nature of the collected material.
Open3DLab operates alongside SFMLab, which hosts models originally associated with Source Filmmaker workflows, and SmutBase, a website dedicated to adult-oriented 3D content.
According to the operator’s breakdown, the crawler downloaded 364 files totaling 54.8GB from 224 SmutBase projects. Including projects marked as adult content on SFMLab, the collection allegedly contained 62.6GB of adult-oriented material across 345 projects.
That represents nearly half of the projects downloaded during the reported collection activity.
Open3DLab says the automated downloads did not appear to distinguish between adult and non-adult content. However, the operator explicitly acknowledges having no evidence that any adult material entered an AI training dataset or became part of a Roblox product.
The allegations are particularly relevant because Roblox has been developing generative AI systems designed to simplify the creation of 3D assets and interactive game environments.
In March 2025, Roblox introduced Cube, its foundation model for 3D and eventually 4D content generation. The technology is intended to help creators generate objects and, ultimately, interactive assets using artificial intelligence.
Roblox has also outlined ambitions for AI systems capable of understanding the geometry, materials, physical properties, and behavior of virtual objects.
These developments form part of a wider shift toward automated game creation tools. Developers throughout the industry are examining how AI can assist with production while retaining creative control, an issue also reflected in Disgaea creator Sohei Niikawa's comments about AI as a development tool rather than a replacement for human storytelling.
Open3DLab believes the crawler may have been gathering data to train or evaluate a 3D foundation model. The operator points to the research-related user-agent identifier and its apparent connection to Roblox’s existing AI work as reasons for that suspicion.
The “trellis2” identifier could also relate to TRELLIS.2, Microsoft’s open 3D generation technology. However, Open3DLab acknowledges that the name alone does not establish which tools were involved or how any collected assets would be used.
The site’s operator says the activity was discovered on October 18, after the initial downloading had stopped. Further requests on October 19 and 20 were blocked by the websites’ servers.
Open3DLab says it contacted Roblox twice seeking clarification but had received no response when its public statement was published.
The operator is asking Roblox to confirm whether it controlled the crawler, explain the purpose of the downloads, clarify whether any assets were used for AI training or evaluation, and address how licensing restrictions and adult content were handled.
It has also requested deletion of the downloaded files and any material derived from them.
Open3DLab says it has preserved the relevant server logs for potential external examination. The complete records have not been independently authenticated in the public reporting reviewed.
For 3D artists and game creators, the dispute highlights unresolved questions about how community-uploaded assets may be collected for generative AI development, particularly when creators specify restrictions on commercial reuse.
For now, the reported 145.5GB download, its attribution to Roblox, and its intended purpose remain based on Open3DLab’s account. No evidence has been made public establishing that the assets were incorporated into Roblox’s AI models.
