OpenAI is hoping to turn ChatGPT into more of an educational toolâand less of an emotional support systemâfor underage users.

The company introduced âChatGPT for Teensâ on Tuesday, a new suite of safety guardrails and features designed to prevent the kinds of abuses that parents and teachers have been flagging for years. The age-specific mode will be turned on automatically for all users who self-identify as being between the ages of thirteen and seventeen, according to OpenAI. The company will also monitor certain indicators, such as the time of day the chatbot is used, to automatically flag underage users trying to fly under the radar. (OpenAI has a minimum age requirement of thirteen years for ChatGPT, but the company has had trouble enforcing that restriction.)

ChatGPT for Teens is being marketed as a healthy middle ground for underage users: âTeens should be able to use AI to learn, create, and explore,â OpenAI wrote in a blog post. âBut that access should come with protections that reflect their developmental stage, reinforce real-world relationships, and support healthy use over time.âÂ

The age-specific mode includes a new âresponsible homework reminder,â for example, which automatically detects when a teenage user is trying to get a quick answer from ChatGPT on a school assignment, and instead helps guide them through the problem step-by-step via the chatbotâs Study Mode feature. Itâs also been designed to withhold sensitive content related to self-harm, eating disorders, sexual content, and other âhigh-risk areas,â according to OpenAI’s blog post.

A less personal (but still customizable) chatbot

One of the big promises of AI, at least according to the companies that have invested billions into building it, is that it will soon allow each of us to have 24/7 access to a personal assistant, health coach, tutor, and even a friend, all rolled into one. The problem is that things can often get much too personal.

Chatbots are trained to imitate human language and maximize user engagementânot truth or personal well-being. Many people, including teens, have as a result come to believe that these automated systems are in fact conscious entities, and formed emotional or even romantic connections with them. Thatâs sometimes had tragic consequences. Last year, OpenAI was sued by the parents of sixteen-year-old Adam Raine, who claim ChatGPT played a role in their son’s decision to take his own life.

Meta, the tech giant behind Facebook and Instagram, has also been at the center of mounting legal controversy: Starting Tuesday, the company will defend itself in court against legal cases raised by attorneys general from Colorado, California, New Jersey, and Kentucky, alleging it knowingly harmed the mental health of underage users. The states are seeking damages of up to $200 billion.Â

OpenAI says its new ChatGPT for Teens mode will make it less likely that underage users will form the kinds of emotional connections with chatbots that have landed the company in hot water in the past. It has updated its model spec to prevent ChatGPT from using âromantic languageâ or suggesting that itâs a feeling, conscious entity, according to the new blog post.

Part of the balancing act that OpenAI is trying to achieve here is making ChatGPT safer for underage users, but not so rigid or boring that theyâll ditch the chatbot for some other tool. To that end, OpenAI says it will still allow some âsimple customization options, like accent colors and voice variations, [to] help make the experience feel personal without blurring the line between a useful tool and a human relationship.â