Roblox 3D Foundation Model Data Scraping Incident

Unauthorized Data Collection of 3D Assets

In October 2026, a crawler identified as trellis2-data-sourcing/1.0 and linked to Roblox infrastructure downloaded 145.5 GB of 3D models from the Open3DLab, SFMLab, and SmutBase platforms. The crawler accessed 1,027 files across 712 projects over a 34-hour period. While the crawler identified itself as part of 3D foundation model research, it ignored robots.txt directives, violated the sites' Terms of Service, and disregarded non-commercial licensing agreements attached to the assets.

The Scope of the Scrape

The scraping activity was broad, failing to distinguish between general-purpose assets and adult-oriented content. Of the 712 projects targeted, 345 were classified as adult content, including 224 files sourced from SmutBase, a site dedicated exclusively to adult material. The operators of the affected sites noted that the crawler's behavior was consistent with automated data collection for machine learning, as the downloaded files included textured, rigged characters and metadata—data types essential for training 3D generative models.

Community and Industry Response

The incident has sparked significant debate regarding the ethics of large-scale AI data harvesting. Critics of the scraping argue that the inclusion of adult content in a dataset intended for a platform used by minors presents a significant safety and oversight failure. Conversely, some observers suggest that such data is often necessary for training safety classifiers to detect and filter adult content, and that public availability of files on the internet inherently invites automated downloading.

Community members and technical observers have highlighted several key perspectives on the incident:

"The internet must understand one simple rule. What’s downloadable will always be downloaded. There are plenty of people who don’t give a flying fk about your LICENSE.md, robots.txt, ToS and whatnot."

"I assume they need the adult content in order to detect adult content generated by their users. I don't see the outrage."

"The entire article points the finger at 'roblox' as an entity, but the behavior seems to indicate it's just an individual doing the scraping - possibly not for corp use. Look at the timelines - this scraper was running on someone's laptop: it turns off when they shut down for the weekend and lunch."

Implications for AI Data Governance

This event highlights the ongoing tension between AI developers and content creators. The affected site operators have formally requested that Roblox clarify the purpose of the data collection, confirm whether the data was used for model training, and commit to deleting the unauthorized copies. As of the report, Roblox has not provided a response to these inquiries. The incident underscores a broader trend where AI labs are increasingly scrutinized for the methods used to acquire training data, often bypassing traditional copyright and licensing norms in favor of bulk automated ingestion.