OpenAI has launched ChatGPT for Teens, a special version of the chatbot for users aged thirteen to seventeen. No one needs to create a new account. Anyone who enters an age within this range during registration will receive the new version automatically. In addition, an age estimation system operates with a single threshold. As soon as it determines that an account is being used by someone under eighteen, it switches the account to the protected version as well. The company is rolling out the new feature to both free and paid personal plans, and the launch will take about two weeks.
Age is estimated based on behavior
A fundamental change takes place before the first message is even written. The age estimation system monitors signals associated with the account, such as the topics of the user's conversations, the time of day they chat, how they use the account, or how long it has existed. The check is also performed for users who entered their date of birth during registration. In practice, age estimation therefore applies to all accounts, since otherwise the company would be unable to identify minors.
Teenagers do not need to provide any proof. An adult whom the system incorrectly identifies can exit the protected mode only by verifying their age through the Persona service. Depending on the country, it requires a real-time selfie, an identity document, or both. According to OpenAI, the company does not receive the document, only information about the date of birth, while Persona deletes the uploaded photograph or ID within seven days.
Study mode instead of completed homework
Educational features are the main attraction of the new offering. Study mode guides students with leading questions and breaks assignments down step by step. New homework prompts can recognize when a student is trying to avoid doing the work and steer them back toward solving it together. Quizzes and clear visualizations of the subject matter have also been added. Study hours then allow a student or parent to set times when study mode starts automatically.
Miami high school mathematics teacher Racquel Gibson notes in OpenAI's press release that not every student has someone at home to help with homework. With well-designed rules, she says, artificial intelligence can fill this gap. The company has also announced a partnership with the organization CodeAI. Its purpose is to explain to students and teachers how models work, how to formulate prompts correctly, and when not to trust them.
The chatbot must not flirt or pretend to have feelings
Safety features are enabled by default. They are based on OpenAI's rules for minors in its model specification and cover self-harm, violence, eating disorders, dangerous challenges, and sexual or graphic content. The updated rules are not limited to blocking romantic and sexualized role-playing. The chatbot must not use romantic language, encourage emotional dependence, or suggest that it has feelings or consciousness of its own.
After ninety minutes of activity within a three-hour period, teenagers receive a reminder to take a break. Before uploading certain images, they are prompted to consider whether they should share sensitive photographs, and the interface repeatedly reminds them that they are communicating with a machine. Parents with a linked account can set so-called quiet hours during which their child cannot access ChatGPT. They also receive alerts in high-risk situations, now including signs of eating disorders. However, parents cannot read the conversations themselves; the system sends only essential information.
What this means for Europe
Age estimation is being introduced worldwide, but in the European Union it will not arrive until the coming weeks because it must comply with local requirements. European users may therefore see some features before automatic behavior-based classification begins operating.
For European adults, this creates a practical inconvenience. The system may make a mistake and place an adult in child mode, after which they must verify their age with an identity document through a third party. Anyone who does not want the system to estimate their age can verify it themselves and thereby disable age estimation. Italy illustrates how requirements differ across the Union. Users there have sixty days to complete verification, otherwise some features will stop working. One small consolation for parents is that the company does not display advertising on accounts with protections for minors.
Critics warn of a false sense of security
The new feature comes as US courts consider lawsuits filed by families. They claim that the chatbot provided information about self-harm or drove young people to suicide. Last September, following a lawsuit over the death of sixteen-year-old Adam Raine, OpenAI announced the development of a system for identifying minors. It has now deployed the system across the board. ChatGPT arrived at the end of 2022 and reached nine hundred million weekly users before receiving protections tailored to teenagers.
Experts warn that the announcement alone solves nothing. Although they say such steps are moving in the right direction, parents should not be lulled into a false sense of security. Some experts argue that AI-based companions are not safe for children because their very design encourages emotional attachment. An open question remains how easily young users will circumvent the new rules, since they have always been very good at overcoming parental restrictions.
Sources: theguardian.com, bbc.com and reuters.com
