Social gaming platform developing new machine learning system for continuous age checks
CategoriesAge Assurance | Biometrics News | Civil / National ID
While it has surely been a busy few months for compliance teams at every large social media company, a special regulatory zeal has been reserved for Roblox, the social gaming platform that has faced numerous accusations of lax online safety measures leading to child sexual exploitation, and numerous lawsuits to match. Regulators globally have summoned Roblox for a word, including, most recently, Thailand’s Ministry of Digital Economy and Society.
A report in Khaosod English says the ministry has had talks with Roblox over stronger safety and child protection measures, “including the possible use of Thailand’s National Digital ID system to verify users’ identities and ages.”
Topics of discussion include age restrictions and age-based filtering, biometric facial age estimation technology, chat moderation systems to detect inappropriate messages, and Roblox’s plan to implement measures based on the International Age Rating Coalition (IARC) rating system in 2027.
Thailand is among a host of Asian nations testing the waters on age assurance legislation for social platforms and beyond. Digital Economy and Society Minister Chaichanok Chidchob has declared online safety a matter of urgency, and says his “efforts are intended to build long-term confidence among users, families and Thai society.”
Chaichanok has raised the possibility of verifying users who wish to access Roblox by linking Thailand’s National Digital ID/KYC system with Roblox user accounts. He is planning a joint working group to study the proposed verification system.
Government-linked systems offer a tradeoff, in that they can provide accurate age verification, but come with tracking and privacy concerns.
‘Age assurance doesn’t end at the selfie’
Roblox currently uses Persona’s facial age estimation technology. However, in its ongoing pushback against reputational and financial catastrophe, it is developing a new modular, multimodal behavioral machine learning system that aims to keep age checks up to date.
“The technology we use for facial age checks was selected because it’s been tested and certified by third-party labs and has proven to be more accurate than self-declared age,” says a statement on Roblox’s website. “This initial age check is necessary, but age assurance doesn’t end at the selfie.”
“To move age assurance beyond a point-in-time signal, we built a high-throughput ML layer that evaluates whether an account’s ongoing behavior still matches its estimated age over time. Our feature-level modular design enables efficient model updates and rapid adaptation to shifting behavioral patterns.”
The motivating idea will be familiar to regular readers of Biometric Update: “no single behavioral signal can reliably capture nuanced behavioral patterns that shift over time and under adversarial pressure.” This could equally apply to fraud prevention, agentic IAM, deepfake detection, and more. Across the sector, the trend is to view online activity as an organic, continuous indicator, which Roblox breaks down into component signal families: social context, experience engagement, account and platform activity, communication patterns and learned user representations.
Behavioral signals and age inference
Which is to say, it’s easier for Roblox to guess a user’s age at any given point in time if it knows who that user is connected to, who they talk to, what games they play and for how long, how old their account is and how often they sign in. Communications patterns yield “model-derived summaries of age-correlated language usage, including how communication patterns vary across context and time.” And higher-level representations capture broader usage patterns; “the same rich representations that predict what a user may engage with next,” Roblox says, “can be post-trained to become predictive of broad age bands.”
It has been clear for some time that most tech companies’ preferred method for age assurance is an internally developed and controlled machine learning algorithm that relies on close analysis of user data to guess at an age profile. There is a plausible argument to be made that social media exists primarily to collect and monetize user data. Even though Roblox is not the same as Instagram, giving any massive tech platform express license to know more about their users is not likely to result in a safer, more private internet in the long run.
Article Topics
age verification | biometric age estimation | facial age estimation (FAE) | gaming | national ID | Roblox | ThaiID | Thailand
