Static

Roblox downloaded 145GB models from us (including adult content) without asking

First reported by Open3dlab ·

The signal ●○○○ Compiled by AI from Open3dlab and Hacker News
Why you might care

Roblox models downloaded from adult sites could be included in AI training data accessible to children.

What happened

On October 15-16, 2026, a crawler identified as "trellis2-data-sourcing/1.0" from Roblox accessed Open3DLab, SFMLab, and SmutBase, downloading 145.5 GB of 3D models from 712 projects. This data included content from adult websites, with 345 of the downloaded projects (62.6 GB) being adult-oriented. The crawler ignored robots.txt and Terms of Service, which prohibit scraping. Open3DLab discovered the activity on October 18th and blocked the crawler, which had resumed scraping on October 19th. Despite attempts to contact Roblox for comment, no response was received. The user agent's email address is linked to a researcher involved in Roblox's "3D foundation model" development, suggesting the downloaded data was for AI training purposes.

What it means

This incident highlights a significant ethical and safety concern: the potential for adult-themed 3D models to enter AI training datasets used by platforms with young user bases. Roblox's automated data collection appears to have lacked any content filtering, raising questions about the safeguards in place for their "3D foundation model" research. The inclusion of adult content, regardless of intent, suggests a critical oversight in their data acquisition process, which could lead to unintended consequences when models generate assets for a platform frequented by minors.

The broad, unchecked scraping of copyrighted and licensed content, including explicit material, underscores a growing tension between AI development and intellectual property rights. This event could spur further legal challenges and demands for greater transparency in AI data sourcing. Creators and platforms are increasingly concerned about unauthorized use of their work, especially when it violates licensing agreements or includes sensitive content that is then indiscriminately collected and potentially incorporated into widely distributed products.

AI-written summary. May contain errors.

Telecom