Teen Mode tries to protect teenagers. In August, OpenAI launched a version of ChatGPT in which every user under 18 will automatically be enrolled. It adds guardrails and learning features designed to reduce harm, but Robin Kirman argues that the system may deprive teenagers of a fundamental need.
The concern follows a lawsuit alleging that ChatGPT contributed to the suicide of Adam Raine, a 16-year-old. Kirman describes Teen Mode as “a tacit admission of culpability” that reveals what the industry now acknowledges—and does not—about its products’ negative effects on young users’ mental health.
Teen Mode provides guardrails against destructive content and sends parent notifications when it detects self-harm concerns. It also prohibits ChatGPT from presenting itself as a friend, claiming personal feelings, or suggesting that it is anything other than a tool.
That restriction responds directly to the Character.AI wrongful death case, in which a fourteen-year-old formed an intense attachment to a chatbot. The lesson is plain enough: an AI system does not need to be a person to occupy an alarming place in someone’s emotional life.
Learning requires the struggle machines remove
Teen Mode’s built-in learning tool, Study Mode, tackles a different risk. Recent research has confirmed that versions of AI that do the work for children can produce short-term grade improvements, but those gains collapse when the tool disappears—and leave students worse off than if they had never used it.
Study Mode aims to produce gains that hold. Leah Belsky, OpenAI’s vice-president for education, calls it a transformation from “a straightforward answer engine into a cognitive tutor”. Its Socratic approach refuses to supply answers, gently pushes back on easy solutions, and creates rather than forecloses the discomfort of not-yet-knowing.
That design is available to any student with a phone. A Socratic companion could offer a real democratization of intellectually rigorous support, surpassing what any human source could provide in scope of knowledge and availability.
Tutors eventually go home. Parents get frustrated, or simply have other things to do. Teen Mode, by contrast, is described as a tireless interlocutor—always available to question, prompt, and redirect the student.
The problem with a perfect intellectual companion
Donald W Winnicott, the British pediatrician and psychoanalyst, spent decades studying how a child’s sense of self emerges. He argued that children need not perfection from their environment, but sufficiency: his most famous concept was the “good-enough” parent.
A “good-enough” parent is good precisely because she is not always good. She is present enough to provide a secure base, but limited enough—distracted, tired, occasionally wrong—that the child must draw on her own resources and discover that she is capable.
A too-good parent would spare the child that struggle entirely, along with the sense of self that struggle builds. Winnicott was writing about early childhood, but the argument extends into adolescence, where independence still has to be practiced rather than announced in a settings menu.
That is where Teen Mode creates its central tension. Its guardrails may prevent destructive exchanges, while Study Mode may stop students from outsourcing their thinking. Yet a system that is always present, always guided, and always ready with a calibrated response can also become too dependable.
“The adolescent whose intellectual life is always accompanied, always guided, always met with a calibrated response, never gets to become her own authority.” The sentence captures the risk beneath the product’s educational promise: support can become substitution, and protection can become another way to remove difficulty.
OpenAI has built Teen Mode to make ChatGPT safer for young users and more useful for learning. The harder question is whether a machine can provide help without quietly taking over the struggle through which teenagers become their own authorities.
Based on
- ChatGPT’s Teen Mode could deprive teenagers of a fundamental need | Robin Kirman — theguardian.com
- Opinion + AI (artificial intelligence) | The Guardian — theguardian.com
- AI chatbots are giving medication advice, and some of it is deadly — thehill.com
- Opinion | The Guardian — theguardian.com
- Today’s AI won’t achieve superintelligence. But catastrophic risk is real | Sunita Verma | The Guardian — theguardian.com
