Roblox kills AI scraper after 145.5GB Open3DLab data pull

Roblox has decommissioned an internal AI agent and pledged that 145.5GB of scraped Open3DLab content — 117.4GB of it adult material — will not be used to train any AI models, a company spokesperson confirmed to GameDeveloper.com.

The admission validates claims posted by Open3DLab administrator "Salaryman," who documented a crawler that hit the 3D asset hub and its adult-content sister sites SmutBase and SFMLab. Roblox says a company researcher built the agent to "collect publicly available 3D assets" and that it "mistakenly" scraped content from Open3DLab because the bot was "failing to operate as intended."

"The agent was decommissioned, and the data was not used for any purpose, including training any AI models," the spokesperson said in a statement. "We are strengthening safeguards for how agents like this are built and reviewed, and will continue to do so as the agentic landscape evolves."

What did the crawler actually take?

The technical details in Salaryman's post deserve attention from anyone building scraping pipelines. The agent "walked through" almost every project across the three hosting sites and downloaded 1,027 files from 712 projects — all user-created assets, many made by 3D artists working in Valve's Source Filmmaker.

The bot retrieved the sites' robots.txt file and, in Salaryman's words, "did not respect" its instructions. It also "ignored" Open3DLab's Terms of Service, which explicitly forbid scraping, and "defied" numerous creators' non-commercial licenses — including licenses attached to models based on other companies' IP, such as Eve from Shift Up's Stellar Blade and Mei from Blizzard's Overwatch.

Open3DLab spotted the crawler on October 18, blocked it and emailed the agent's creator, identified through an email address in the agent's own data. The bot resumed scraping on October 19, evading the operators' countermeasures. Salaryman identified the creator as a Roblox researcher who co-authored the paper defining Roblox's core generative AI system, Cube. Other data signatures marked the agent as a collector assembling training datasets.

Why adult assets were the real problem

Before Roblox's confirmation, the stakes were straightforward: adult training data flowing into a generative model serving Roblox users would put any user who triggered it in violation of Roblox's own terms of service.

"Models reproduce what they learn from," Salaryman wrote. "If adult training data reaches a model that generates assets for Roblox, the only thing between that data and children is whatever filter Roblox builds afterwards, because its data collection didn't filter anything."

He was clear that Open3DLab doesn't believe Roblox set out to harvest adult content. "But that is exactly the problem: as far as our logs show, nobody checked," he wrote. "A crawler working for a company whose platform tens of millions of children use was pointed at an adult site, and it took explicit models without telling them apart from anything else."

The episode lands on fertile ground. Roblox has spent 2026 aggressively scaling moderation — so aggressively that investment research firm Bleecker Street Research estimates the crackdown negatively impacted user growth. The company remains in litigation with government entities and parents of exploited minors over allegations it enabled harassment and exploitation, claims Roblox has repeatedly denied. CEO David Baszucki has drawn criticism for pushing back on child-safety questions in multiple interviews.

Open3DLab, for its part, isn't staking out an anti-adult-content position. Adult material is "our bread and butter," Salaryman noted: "We think consenting adults should be able to access and enjoy this content freely, without the government looking over their shoulder."

The objection is narrower and sharper: "What we object to is our adult sites being emptied into an exploitative gaming company's training data without anyone asking, and being mixed into content offered to literal children."

For studios building agentic data-collection tools, the lesson is operational, not moral. An autonomous crawler that reads robots.txt and ignores it, resumes after a block, and outlives its welcome by a full day is a liability — especially when the company doing the scraping runs a platform for children. Watch whether Roblox publishes details of those new agent safeguards, and whether the litigation record forces the issue first.