OpenAI limits teens to dedicated version of ChatGPT promising more safeguards

OpenAI will restrict teenagers to a new version of ChatGPT with greater mental health safeguards and features to encourage critical thinking as it seeks to respond to a backlash over the effects of AI on young people.

The $852bn start-up will roll out the dedicated ChatGPT for 13- to 17-year-olds globally starting on Tuesday, with stricter controls around image generation and content regarding self-harm, violence, eating disorders and sex.

The move comes as OpenAI faces multiple lawsuits from families who claim their children were harmed by using ChatGPT as it gears up for a potentially record-breaking initial public offering within the next year.

Scrutiny of AI’s effects on young people comes as several countries, including the UK, move towards bans on social media for under-16s — measures that typically exclude AI chatbots but have heightened the pressure on tech companies.

OpenAI has been sued by the family of Adam Raine, a 16-year-old from California who died by suicide after discussing ending his life with ChatGPT, which provided detailed instructions, according to court documents.

Families whose children were killed in the mass shooting at a school in Tumbler Ridge, Canada, in February have also sued OpenAI, after it emerged that the teenage shooter had discussed violent intentions with ChatGPT in the months leading up to the attack. The account was banned months before, but the company did not contact law enforcement.

“Those scenarios are heartbreaking . . . One incident is too many,” said Lauren Jonas, head of youth and families at OpenAI.

“Safety work is work that’s never done . . . You can see the bags under my eyes,” Jonas added in an interview. “We are constantly learning and improving around how we identify when a teen is having a hard time and serve them the right resources that they need.”

ChatGPT has parental controls which send safety notifications in high-risk situations, such as if a child is discussing suicide or self-harm. The new version for teens adds notifications on eating disorders and violence.

“It is a balance between a teen’s right to privacy and [the] severity of the harm itself,” Jonas said, addressing why these issues were only added in the latest update.

OpenAI’s policy bars children under 13 from the chatbot and requires parental consent for users under 18. It will use a user’s stated age alongside “a combination of signals” about their usage to determine when to present the teen version.

ChatGPT for teens will steer them towards educational applications and include more frequent break reminders and “quiet hours”, times that users or parents can set when ChatGPT is blocked.

Jonas said OpenAI “did not see a problematic-use problem right now or an extended-use problem on ChatGPT”. But because of the lack of research on the subject, the company has “taken a conservative approach”.

OpenAI is also trying to address growing concern that chatbots giving instant answers to teens’ questions harm their critical thinking and learning through a dynamic known as “cognitive offloading”.

The company said the new version will have an updated ‘study mode’ that guides teens through problems step by step, rather than simply providing the answer.

Jonas said 90 per cent of teens on ChatGPT use it for learning and OpenAI “want to continue to build on this experience and make a learning tool”.

“That is the goal more than anything else is to make it a learning tool,” she added.