Roblox confirms no adult content used in AI training after scraping Open3DLab data
By Image by Photo Agency via Adobe Stock
Roblox Corporation has acknowledged claims shared by 3D asset platform Open3DLab that one of its AI agents acquired 145.5 GB of materials such as artistic assets and 3D models, including adult-themed content found on its subsidiary platforms Smutlab and SFMLab. The organization is now informing the public that the AI agent has been shut down, and the collected data will not be employed in the training of any AI models.
A representative from the company confirmed the update in a statement given to GameDeveloper.com. The spokesperson validated the assertions made by Open3D admin “Salaryman” that a Roblox researcher developed an AI agent to “gather publicly accessible 3D assets,” which “accidentally” scraped materials from Open3DLab. The bot was “not functioning as designed,” they noted.
“The agent has been terminated, and the data has not been utilized for any objectives, including the training of any AI models. We are enhancing safeguards regarding how agents like this are developed and evaluated, and we will keep doing so as the agentic environment advances.”
Related:Rockstar Games employees to cast their votes on unionization just as Grand Theft Auto VI releases
Developers and industry professionals curious about the technical methodology utilized by the Roblox bot will find much to consider in Salaryman’s remarks. Evidently, the crawler “navigated through” nearly every project across the three hosting platforms, and downloaded “1,027 files from 712 projects.” These files were all created and uploaded by Open3DLab contributors, which consist of 3D artists and creatives using Valve’s Source Filmmaker software, many of whom regard Smutbase and SFMLab as reliable sources for adult materials.
According to Salaryman, the AI agent downloaded 117.4 GB of adult-themed content from SmutBase and SFMLab. This occurred after the bot accessed the site’s robots.txt file, which provides guidelines for web crawlers. It “did not adhere to” the instructions in that file and “disregarded” Open3DLabs’ Terms of Service, which prohibit scraping. He also mentioned that the bot “violated” many creators’ non-commercial licenses, many of which are applied to models based on other entities’ intellectual property.
