Safer AI for teens – but at what cost? Age assurance and the role of civil society

This week OpenAI launched a dedicated ChatGPT for Teens (ages 13–17) with stricter limits on harmful content, a study mode and parental controls. The service automatically estimates whether a user is under 18 and routes them into the safer version. It is part of a larger movement: the internet is becoming age-tiered, driven by new age-assurance requirements around the world.

For nonprofits – especially those working with children, youth, schools or digital safety – this is a development worth understanding and having a view on.

The good intention

Let us start with the positive: young people have used AI tools without protection for years. Building safer boundaries for the youngest – with limits on self-harm, eating disorders and sexual content – is a good thing, and something children’s and youth organisations have long called for. That the world’s largest AI companies are now taking children’s safety seriously is a win for work many in the sector have carried.

But be aware of the downside

Here civil society has an important watchdog role. To protect minors, services must in practice assess the age of all users – either by requiring ID, or by "guessing" age from behaviour, language or face. Both carry a privacy cost. Measures meant to protect children can therefore end up introducing surveillance and identity checks of the whole population. That is a trade-off that should happen openly, not hidden in terms no one reads.

Three questions organisations should ask

• Is the age assessment accurate? Estimates get it wrong sometimes – both by letting minors in and by locking out adults, or vulnerable people who do not want to identify themselves.
• What happens to the data? Information collected to protect children is still sensitive and can leak or be misused.
• Does it hit the vulnerable? Requirements to prove age can shut out precisely those who most need anonymity – for example young people seeking help they do not want parents to see.

What does it mean for your organisation?

• For work with children and youth: safer AI tools for teens can be an asset in your work – but always judge for yourselves whether the content and boundaries are suitable, rather than blindly trusting a "teen mode."
• Use your voice. Nonprofits have the credibility to demand that age assurance is designed to be privacy-friendly and accurate, and that the needs of the vulnerable are safeguarded. That is a debate you should be part of.
• Teach critical thinking and safe use. However good the boundaries become, the best protection is still young people who learn to use the technology wisely. This connects to the reflections on AI and children we wrote about recently.

The conclusion is balanced: protecting children online is right and important, and this week the tech companies took a step many have waited for. But the way it is done – by services assessing everyone’s age – is a big change with real privacy costs. For a sector that both protects children and defends individual rights, this is a case where both concerns must be held up at once.