Watchdog Calls ChatGPT for Teens an Unacceptable Risk

Common Sense Media, a nonprofit that evaluates technology for youth safety, has labeled OpenAI's ChatGPT for Teens an unacceptable risk. According to The Verge, the group found the product fails to notify parents when it should, offers inadequate help in crisis moments, and still completes kids' homework despite claims to the contrary.

The finding matters because ChatGPT for Teens was itself the mitigation. OpenAI built a teen-specific mode to answer mounting pressure over how chatbots handle minors. The watchdog is now saying the fix does not do the thing it was built to do. That is a different and more serious claim than the usual complaint that a product could be safer.

Key points

• Common Sense Media rated ChatGPT for Teens an unacceptable risk, its most severe category.
• The product reportedly fails to alert parents in situations where it is supposed to.
• Crisis-response handling was judged inadequate, the highest-stakes failure mode.
• It still completes homework, undercutting one of the stated academic guardrails.

Stated guardrail          Watchdog finding
────────────────          ────────────────
parent alerts        →    did not fire when expected
crisis intervention  →    inadequate support given
no homework answers  →    still completes assignments
        │                        │
        ↓                        ↓
  marketed as safe        gaps at the worst moments

The crisis-support failure is the one to sit with. A teenager reaching a chatbot in a bad moment is exactly the scenario these guardrails are supposed to cover, and it is the scenario where a weak response does the most damage. Parent notifications are the backstop for that same risk. If both are unreliable, the safety story reduces to a mode that mostly affects tone, not outcomes.

There is a structural problem under this. Age gates and teen modes assume the system can reliably detect context: who is using it, how old they are, whether a conversation has turned dangerous. Large language models are not built to make those judgments dependably, and bolting a safety layer onto a general-purpose chatbot does not give it that capacity. The gap between what a teen mode promises and what the underlying model can detect is where every one of these failures lives.

The pressure is not going away. Regulators and attorneys general are already scrutinizing how AI chatbots treat minors, and a third-party failing grade on a flagship product is the kind of evidence that invites rules. OpenAI can dispute the methodology, and it may. But the burden has shifted: a company that shipped a teen product as proof of responsibility now has to show the product performs when it counts, not just that it exists.