Discord is reintroducing age verification measures, this time employing an "age estimation model" designed to infer user age based on account activity rather than direct personal data. This approach aims to comply with legal obligations while minimizing user privacy intrusion.
The system analyzes patterns in account behavior, such as server connections, general activity levels across channels and servers, and account age. Discord stated that this model "does not look at your messages, calls, or other content." It posits that the usage patterns of teenagers differ from those of adults, and its model has learned to distinguish these differences.
According to Discord, this new model is intended to be less intrusive than previous proposals, which had faced significant backlash. The company stated, "We believe meeting age requirements shouldn’t require ID checks or selfies for everyone. We'd prefer to give you more choices, including less intrusive options that don't require biometrics or sensitive personal data."
Discord anticipates that this estimation model will mean "more than 90% of you will not be asked to confirm your age."
This marks Discord's second attempt at implementing a broad age verification system. The company initially announced plans in February for a system that would have locked users into a "teen-appropriate experience" if they did not verify their age. That rollout was tied to Persona, a vendor linked to Palantir co-founder Peter Thiel. Following widespread criticism and privacy concerns, Discord delayed the rollout and severed ties with Persona, stating it did not meet their privacy standards.
The current iteration seeks to avoid these pitfalls by relying on an AI model that analyzes existing account data, aiming for a balance between regulatory compliance and user privacy. However, concerns remain about the accuracy and implications of such predictive age estimation systems, especially in light of recent online safety legislation in regions like the UK, which have sometimes led to easily circumvented verification methods.