YouTube will begin introducing a new artificial intelligence (AI) and machine learning (ML) age estimation system in the United States from 13 August, designed to identify users who are under 18, even if they entered a false date of birth when creating their account.
Unlike the current system, which relies on the date of birth provided during sign-up, this new model will look at different activity signals to estimate a user’s age. These signals include the kinds of videos watched, the topics searched for on the platform, and the length of time the account has been active. By doing so, YouTube aims to catch cases where minors are using adult accounts or have lied about their age.
Restrictions for flagged accounts
If the system decides an account likely belongs to someone under 18, YouTube will introduce a series of restrictions. Personalised advertising will be switched off, digital wellbeing features, such as bedtime reminders and screen time tracking, will be enabled, and certain types of content will be limited, especially where repetitive viewing may be harmful.
The company says the measure is part of its ongoing effort to protect teenagers and children from unsuitable material and data collection for targeted advertising. However, YouTube admits that the AI model is not flawless. Adults may be incorrectly flagged as minors, for example, parents who regularly watch children’s content could trigger the system.
Gradual global rollout planned
If this happens, the account holder will need to verify their age to remove the restrictions. Verification options include submitting a government-issued ID, uploading a selfie for comparison, or providing credit card details. While these methods confirm identity, they may raise privacy concerns for users reluctant to share sensitive information with the platform.
YouTube plans to monitor how well the system performs in the US before extending it to other countries. The company says further refinements will be made based on user feedback and accuracy rates before any wider rollout.
