Open3DLab says Roblox-linked crawler took 145.5 GB of assets

By Sophie Lindqvist

A crawler using a Roblox email address downloaded 145.5 GB from three 3D-model sites, according to logs reviewed by independent community Open3DLab, which has received no explanation from the gaming platform.

The haul contained 1,027 files from 712 projects. Open3DLab believes the files were gathered for AI training, although its report says that purpose cannot be proved.

What did the crawler collect?

Open3DLab said the crawler mapped "almost every project" on Open3DLab, SFMLab and SmutBase. It read what the community called the practically entire catalogue of 18,926 projects.

The run covered 34 hours and 26,429 requests before the crawler suspended operations.

• Total download: 145.5 GB, comprising 1,027 files from 712 projects.
• SmutBase: 364 files from 224 projects, totaling 54.8 GB. The sister site focuses on adult content.
• SFMLab: 345 adult-marked projects, accounting for 62.6 GB.

The crawler ignored robots.txt and the terms of service, Open3DLab said. It also bypassed creator licenses prohibiting noncommercial use.

The report described the downloaded material as complete copies. Those failures matter independently of the crawler's claimed research purpose.

The crawler identified itself as trellis2-data-sourcing/1.0, described its work as 3D foundation model research, used a redacted @roblox.com address, and routed every request through a single Amazon AWS address.

Those identifiers link the crawler to Roblox. They do not independently establish who controlled it or what instructions it received.

How long did the scraping continue, and why did it stop?

Open3DLab staff detected the request surge on October 18. They blocked it by the following morning.

Further attempts arrived on October 19, neatly within office hours and with only a one-hour pause around lunchtime. The crawler did not reduce its request rate after the block, according to Open3DLab.

A final, tapering attempt came on October 20, likely to test rate limits, the community said. Requests stopped later that day.

Open3DLab said its free download servers enforced speed and concurrent-download caps, which the community described as the crawler's main limiting factor rather than any respect for site restrictions.

Open3DLab requested comment from Roblox. The company had not replied when the report was published.

Open3DLab also says it has changed its setup to reduce recurrence risk. No specific configuration changes are disclosed, so their effectiveness cannot be assessed from the available record.

Is AI training the proven purpose?

Roblox has been building a 3D foundation model since March 2025, when it introduced Roblox Cube, its core generative AI system for 3D modeling and character rigging, with characters and avatars listed as future goals.

Roblox also beta-launched 4D generation in February 2026. The technology can build in-game models from natural-language prompts.

Open3DLab says the crawler's email address matched one researcher listed on the Roblox Cube research paper. It withheld the researcher's name.

That match strengthens the link to Roblox's model work. The logs still do not reveal what the researcher or the company intended.

Open3DLab called the collection of training or evaluation data its "most likely explanation." It also stated that it could not prove the theory, and Roblox has not supplied an explanation.

Why did adult content enter the download?

Open3DLab does not think Roblox deliberately targeted adult material, yet its logs indicated that nobody checked the crawl, which the community called a poor governance position for a company serving tens of millions of users.

Roblox has over 33 million users and billions of game experiences. Forty percent of players are reportedly under 12.

The distinction matters because the crawl crossed separate sites and content classes without evident human review. Open3DLab's complaint therefore concerns both data collection and the apparent absence of checks.

The competitive stakes are straightforward: foundation models need training and evaluation data, while creators and adult-content sites need enforceable limits on reuse. Unless Roblox identifies the crawler's operator and explains the missing checks, its 3D-model roadmap will remain shadowed by the 145.5 GB download.