Roblox accused of scraping 145GB in 3D models for AI
Published on · Updated on
Independent community Open3DLab accuses a Roblox crawler of scraping 145GB in 3D assets without consent, sweeping up non-commercial and explicit files.
Roblox Corporation found itself at the center of a major controversy in the 3D computer graphics and modding community in late October 2026. The independent 3D modeling platform Open3DLab published a comprehensive technical report revealing that an automated web crawler directly tied to the gaming company conducted an unauthorized scraping operation across its infrastructure. Over 34 hours of continuous activity, the AI data-sourcing agent executed more than 26,000 requests and downloaded 145.5 gigabytes of 3D assets, sweeping up hundreds of projects protected by non-commercial licenses along with a substantial volume of explicit adult content.
The revelation raises urgent ethical and regulatory concerns for the gaming giant, whose player base consists predominantly of children and teenagers. As tech firms accelerate investments into generative AI capable of synthesizing real-time 3D meshes and environments, indiscriminate data collection on the open web has once again laid bare the risks of unchecked automation and inadequate human oversight in corporate training pipelines.
Massive data harvesting: how the crawler downloaded over 145 GB
According to technical data published by the administrators of Open3DLab and its sister sites, SFMLab and SmutBase, the anomalous traffic was first detected on Sunday, October 18, 2026. System administrators noted extraordinary load spikes on their free download servers originating from a single Amazon Web Services (AWS) IP address.
An audit of server access logs revealed the full scope of the operation: the crawler indexed 18,926 project landing pages and fired off 26,429 HTTP requests across roughly 34 hours. By the time it was halted, the bot had gathered complete copies of 1,027 files across 712 separate community projects, amounting to 145.5 GB of digital assets. The scrape was only limited because the community repositories enforce automated bandwidth caps and concurrency limits on free downloads.
Rather than merely indexing metadata or descriptions, the crawler downloaded full compressed packages containing 3D polygonal geometry, skeletal rigs, high-resolution textures, and rendering project files. This behavior strongly indicated that the harvested data was structured specifically for assembling a large-scale training corpus for 3D generative foundation models.
User-Agent transparency and the link to 3D foundation model research
The definitive link connecting the automated sweep to Roblox was found directly in the connection headers. Across thousands of requests, the crawler consistently broadcasted the following User-Agent string:
“trellis2-data-sourcing/1.0 (3D foundation model research; [redacted]@roblox.com)”
The string explicitly included an official Roblox corporate email address and cited 3D foundation model research. The naming reference to trellis is closely aligned with recent breakthroughs in deep learning and geometric diffusion models designed to synthesize complex 3D meshes from text prompts or single images—a field where Roblox has been actively researching tools to let users generate in-game objects effortlessly.
Community investigations quickly determined that the associated email address belonged to a research scientist credited with academic publications on 3D generative diffusion architectures. Despite displaying transparency in its header, the automated agent bypassed the platform's robots.txt instructions and breached the website's Terms of Service, which strictly prohibit automated data scraping.
The adult content controversy: 62 GB in explicit material
The most alarming dimension of the incident involves the nature of the assets ingested. According to Open3DLab's technical breakdown, 345 of the 712 downloaded projects contained explicit adult or NSFW (Not Safe For Work) content, representing roughly 62.6 GB of harvested data. Within that volume, 364 files totaling 54.8 GB were obtained directly from SmutBase, an associated portal dedicated exclusively to adult 3D art.
The files gathered without filtering spanned fan-made models of mainstream video game characters such as Mei from Overwatch and Eve from Stellar Blade, alongside completely unrated explicit models. The indiscriminate sweep proved that the data collection agent operated without basic semantic filtering or age-rating safeguards.
In its public blog post, the Open3DLab team offered a measured assessment of the findings:
“We don’t think Roblox set out to collect adult content, and we have no evidence that any of it reached a model or a product. However, our logs suggested no one had checked the material being collected.”
For gaming and cybersecurity analysts, the reality that a corporate crawler sucked up gigabytes of adult content exposes severe vulnerabilities in automated research pipelines. With Roblox continuously facing regulatory scrutiny over child safety across its global platform—which hosts popular family experiences like Yakuza Blocks on Roblox—any association between sensitive files and internal AI systems creates immediate public and legal fallout.
License violations and disregard for community creators
Beyond the sensitive content, independent creators expressed deep outrage over intellectual property disregard. The community audit verified that 423 of the 712 scraped projects were released under explicit Non-Commercial licenses, set by original authors to prevent commercial corporations from exploiting their creative efforts without consent.
Furthermore, the crawler accessed the platform's robots.txt file before proceeding to systematically ignore its directives. Open3DLab's terms explicitly forbid web scraping, a policy instituted to safeguard server bandwidth and protect thousands of independent digital artists from unauthorized mass extraction.
| Metric / Dimension | Audit Data Recorded |
|---|---|
| Total data volume downloaded | 145.5 GB (1,027 files across 712 projects) |
| Adult / NSFW content volume | 62.6 GB (345 projects, 54.8 GB from SmutBase) |
| User-Agent identifier | trellis2-data-sourcing/1.0 (3D foundation model research) |
| Non-commercial licenses breached | 423 community projects under Non-Commercial terms |
| Scraping duration and request count | 34 hours across 26,429 HTTP requests |
| Current operational status | Agent decommissioned by Roblox; data deleted |
Roblox official response: agent decommissioned and data deleted
Following public reporting led by News18 and extensive community debate, Roblox addressed the allegations through an official corporate spokesperson. The company acknowledged that a researcher created an automated AI agent intended to gather public 3D assets, but admitted the system malfunctioned and strayed beyond its intended operational parameters.
Roblox stated that the agent mistakenly scraped content from Open3DLab and failed to operate as intended. The company confirmed that the automated crawler has been permanently decommissioned and asserted unequivocally that none of the harvested files were ever utilized to train generative models or incorporated into any live product. Furthermore, the company affirmed that the downloaded materials have been purged.
Roblox stated it is reinforcing internal governance protocols and review mechanisms to ensure experimental research agents comply strictly with external terms and copyright boundaries. Nonetheless, 3D artists and modding enthusiasts remain skeptical, noting that claims of rogue bots and unintended automation have become a standard defense in corporate data-gathering incidents.
Generative 3D AI and the boundaries of ethical web data
The clash between Roblox and Open3DLab highlights one of the most critical debates confronting the current wave of artificial intelligence: what boundaries must protect digital creators as tech giants scour the internet for training fuel? For readers tracking the latest developments across our gaming category hub at O Santuário Geek, the race to build real-time 3D synthesis tools remains fiercely competitive, yet shortcut methods that ignore site policies face fierce community resistance.
Digital artists are increasingly demanding enforceable opt-out standards, technical protection measures, and genuine accountability from AI labs. As major entertainment studios strive to balance rapid feature deployment with ethical integrity, the Roblox scraping episode serves as a powerful cautionary tale about the indispensable necessity of human oversight in automated AI engineering.
Suggested and related articles
Unbelievable! Argentine peso sinks so low that Roblox money is now worth more
Anime Dice codes in Roblox grant free lucky spins and gems to unlock the rarest heroes free
Fisch Codes in Roblox: Updated List With Rare Boats and Thousands of Free Coins to Boost Your Fishing
Complete guide to all Search for the Needle classes in Roblox with spin odds and abilities to farm fast
The Veil Codes in Roblox Unlock Dark Secrets and Grant Free Race Spins
