Teen AI planning needs verifiable hand-offs, not another persistent helper
OpenAI announced college-planning and study features for ChatGPT for Teens as an independent assessment reported safeguard failures. Education buyers need to test when the system hands control to a student, parent or professional.
What happened
OpenAI announced College Planner, flashcards and expanded study tools on 7 October. The same day, Common Sense Media published a risk assessment based on more than 4,000 prompts and rated ChatGPT for Teens an unacceptable risk.
Why it matters
Planning and learning support can become dependency or false reassurance when users cannot see the boundary between guidance, verified requirements and situations requiring trusted human help.
OpenAI announced new teen learning and planning features on 7 October, including a College Planner intended to organise requirements, deadlines and financial-aid steps. The company also cited product-use counts and a planned student advisory programme. The same day, Common Sense Media published a risk assessment based on more than 4,000 prompts run before and after teen mode launched. It reported failures in crisis support, parent alerts, anthropomorphic responses and homework boundaries. AP reported OpenAI’s response that much of the testing may have preceded full parental-control activation.
The evidence is contested and does not establish incidence among ordinary users. It does establish a testable disagreement about control behaviour.
Test the hand-off boundary
Before a school or family relies on planning features, create scenarios for outdated requirements, conflicting deadlines, financial questions, distress, requests to bypass study mode and repeated dependency cues. Record whether the system cites an authoritative source, marks uncertainty, pauses, escalates or hands control back.
Separate task support from welfare safeguards. Completion reminders are not evidence that crisis detection or learning transfer works. Ask students to complete a matched planning or learning task without the assistant and explain the decision in their own words.
The immediate governance decision is not whether every teen should use or avoid one product. It is whether the deployment has observable, independently retestable hand-offs to students, caregivers, educators and qualified support. Human review remains essential for high-stakes education, financial-aid and wellbeing decisions.
Publish the test protocol and version identifiers so an external reviewer can distinguish a fixed failure from a changed prompt set. Safeguard evaluation should include false escalations as well as missed escalations, because excessive alerts can train families to ignore the channel. Document who receives an alert, what evidence accompanies it and what happens when no responsible adult is reachable.