Roblox-Linked Crawler Scrapes 145 GB of Community 3D Data

The automated process ignored site restrictions to harvest thousands of files, raising concerns over unauthorized AI training.

Updated on Oct. 27, 2026 in Artificial Intelligence

Isometric editorial illustration of a complex, abstract digital lattice of 3D wireframe polygons, representing unauthorized large-scale data harvesting. — A scraper using Roblox-affiliated credentials extracted 145 GB of 3D data from community sites, fueling concerns over unauthorized AI training practices. AI Illustration. Upload story photo >

Live Poll

Should companies be prohibited from using non-commercial content to train AI models without explicit permission?

An automated crawler using Roblox-affiliated credentials recently scraped 145.5 GB of 3D model files from three community websites. While site operators suspect the data was collected for AI model training, the crawler's true origin and purpose remain unconfirmed by Roblox.

Why it matters

This incident highlights the ongoing tension between AI developers seeking large-scale training datasets and independent creators managing intellectual property. It underscores the potential for unauthorized data acquisition to bypass both robots.txt protocols and specific licensing restrictions.

The crawler performed 26,429 requests over 34 hours to scrape 712 projects. Of these, 423 files were restricted to non-commercial use, and 345 contained adult content.

The players

Roblox

A gaming platform and social engine that hosts user-generated 3D content and manages a large-scale ecosystem of digital assets.

The details

The scraper identified itself via the user agent 'trellis2-data-sourcing/1.0', which included a Roblox email address. It originated from a single AWS address and systematically ignored robots.txt files—a standard text file instructing search bots which pages to avoid—and site terms of service. By bypassing these access controls, the process automated the extraction of 3D assets directly from Open3DLab, SFMLab, and SmutBase.

October 15-16, 2026: Crawler scrapes 1,027 files from three websites.
October 18, 2026: Site operators discover the scraper and implement blocks.
October 19, 2026: The crawler returns and is blocked by operators again.
October 20, 2026: Crawler activity ceases at 15:00.
October 26, 2026: Operators publish a first-hand account of the scraping incident.

The Tech Race

This scraping event mirrors the broader industry trend of aggressive data acquisition for AI model training. It suggests that independent 3D content repositories are increasingly vulnerable to automated ingestion efforts similar to the battles seen in text and image datasets.

Independent creators hosting assets on these platforms may find their content used for model training without authorization or compensation. As automated scraping becomes more efficient, creators must rely on increasingly strict server-side blocking to protect proprietary files.

The takeaway

The lack of response from the platform highlights the difficulty in holding large developers accountable for automated data gathering. Watch for official statements from the platforms involved or further technical audits regarding the crawler's infrastructure.