Beyond the Selfie: How Roblox’s Age-Assurance System Helps Keep Age Checks Current

This post details the evolution of Roblox's age-assurance system by introducing a high-throughput ML layer that continuously evaluates an account's ongoing behavior against its estimated age, moving beyond the initial point-in-time facial age estimation. It describes the system's modular, feature-level design for efficient model updates and adaptation to behavioral shifts. The system processes nearly two billion account-level records daily, utilizing diverse ML models (gradient-boosted decision trees, multilingual transformers, neural networks) and behavioral signals (social context, experience engagement, account activity, communication patterns). A gradient-boosted meta-learner combines submodel outputs into a final calibrated estimate for ongoing age assurance.