On January 20, 2026, OpenAI announced the rollout of an age prediction system for ChatGPT consumer plans designed to help determine whether an account is likely to belong to a user under the age of 18.

Rather than relying solely on a user's stated age, the system considers multiple signals, including how long the account has existed, typical times of day when ChatGPT is used, patterns of use over time, and the user's stated age. When the system is not confident about a user's age, OpenAI says it will default to the safer experience intended for younger users.

    Accounts predicted to belong to users under 18 receive additional protections designed to reduce exposure to potentially inappropriate material, including graphic violence, self-harm, sexual, romantic or violent role-play, dangerous challenges, and extreme beauty standards or unhealthy dieting content.

Adults who are incorrectly placed in the under-18 experience can verify their age and regain the standard adult experience. OpenAI says it will continue to improve the accuracy of its age-prediction system.

Comment from GAME-F

For child-safety measures to be effective, a service must first be able to determine who is actually a child. In this respect, age prediction that does not rely solely on self-reported information is an important experiment.

    Traditional age restrictions often depend on users simply entering their date of birth, leaving open the possibility that registered information does not reflect their actual age. Combining signals such as usage times and longer-term patterns offers one technological response to this problem. At the same time, analyzing user behavior to estimate age creates a new question about the balance between child safety and privacy. False classifications also matter: adults may be identified as minors, while minors may potentially be classified as adults.

    We are moving from an era of “asking a user's age” toward one of “predicting a user's age.” What matters is not simply how sophisticated the technology becomes, but whether it actually makes children safer—and how much analysis of user behavior should be considered acceptable in achieving that goal.

References

OpenAI — Our approach to age prediction, January 20, 2026