Roblox Implements Real-Time Multimodal Moderation to Catch Contextual Violations
The shift from reviewing individual items to evaluating entire scenes in real time represents a technical change in how Roblox handles moderation, targeting contextual violations that previous tools missed.
Reporting from 1 sources: 4Gamer.net.
Roblox has introduced a real-time multimodal moderation system that evaluates entire scenes-including 3D objects, avatars, and text-simultaneously, rather than reviewing individual items separately. The system is designed to catch violations that occur only when harmless elements are combined, such as creating inappropriate figures or outfits. Naren Koneru, head of Trust & Safety Engineering, explained the mechanism in an interview with 4Gamer.
Roblox has deployed a new automated moderation system that evaluates entire scenes in real time, including 3D objects, avatars, and text, rather than reviewing individual items separately. Naren Koneru, Vice President of Engineering for Trust and Safety, told 4Gamer that the previous tools missed contextual violations that occur when harmless items are combined-for example, combining avatar clothing items to create outfits that suggest prohibited figures. The system uses an AI model to determine whether the combination of elements at a given moment violates community standards.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.