We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
OpenAI Blog·Jan 20·5 min read
Our approach to age prediction
TL;DR
ChatGPT lacked reliable age verification, creating risk of exposing minors to inappropriate content. OpenAI built an age prediction system that classifies accounts as under or over 18 using account signals and behavioral patterns.
✦ Why It Matters
Engineers can implement age-gated safeguards without explicit user verification, protecting minors while preserving platform access.
Key Takeaways
- Use behavioral signals and account metadata to infer user age without explicit submission.
- Deploy age-specific safeguards rather than binary allow/block decisions for minors.
- Plan for continuous accuracy improvement through iterative retraining and feedback loops.
- Test age classifiers incrementally to validate real-world performance before full rollout.
Full Summary
Online platforms must protect minors from age-inappropriate content, but determining user age without explicit verification is difficult. OpenAI developed an age prediction classifier for ChatGPT that estimates whether an account belongs to a user under or over 18 years old.
The system analyzes account metadata and usage patterns rather than requiring manual age submission. This approach allows OpenAI to apply differentiated safeguards—such as content restrictions or feature limitations—specifically to teen accounts without blocking access entirely.
The classifier is being rolled out incrementally, with accuracy refined over time through feedback and additional training data. Results show the system can meaningfully reduce exposure of minors to harmful content while maintaining usability for younger users.