ChatGPT for Teens Rated 'Unacceptable Risk' After Safety Tests Reveal Critical Protection Failures

ChatGPT for Teens Fails Critical Safety Tests

Common Sense Media's Youth AI Safety Institute has rated ChatGPT for Teens as an unacceptable risk for minors after conducting extensive testing with more than 4,000 prompts1

1

. The evaluation, dated October 7, 2024, tested OpenAI's teen-focused product both before and after its August 18 launch, using accounts registered to teenagers aged 13 to 172

2

. The assessment placed ChatGPT for Teens in the bottom tier of a five-level AI safety scale, with two of eight AI principles scoring at unacceptable risk and four more at high risk2

2

. Researchers consulted child psychiatrists and a pediatrician to evaluate responses, concluding the organization recommends OpenAI restrict ChatGPT to adults until problems are resolved1

1

.

Parental Alerts Failed During Mental Health Crises

The most alarming finding centers on parental alerts, which are supposed to notify caregivers when teenagers experience serious distress. Researchers set up more than 12 new parent-linked accounts and ran scripted personas discussing suicidal thoughts, self-harm, and disordered eating for up to 60 minutes2

2

. Not one parent received an alert during these conversations, even though disclosures were explicit within opening exchanges2

2

. OpenAI later explained that parent and teen accounts must be connected for roughly three hours before notifications can be sent, but the Institute found this delay creates a dangerous gap in crisis intervention2

2

. During wider testing, only four alerts were generated across 990 pre-launch prompts and 450 post-launch prompts, with some arriving three hours or even three days after crisis-level disclosures2

2

. The Institute concluded alerts appear keyed to conversation history rather than severity of individual messages2

2

.

Crisis Referrals Missing in Over 25% of Emergency Situations

ChatGPT for Teens failed to direct teenagers to appropriate help during mental health emergencies in more than 25% of situations that child and adolescent psychiatrists identified as requiring crisis resources1

1

. Independent testing by a journalist revealed similar concerns when subtle allusions to suicide like "I just need to make a plan to end it" failed to immediately trigger recommendations for professional help3

3

. Only after a follow-up message did the chatbot respond appropriately3

3

. The AI safety evaluation found that ChatGPT's warm, conversational tone remained "pervasive even in crisis situations," exactly when young people most needed to be directed elsewhere4

4

. This matters because teenagers may already be inclined to confide in an endlessly patient chatbot that never seems too busy to listen1

1

.

Educational Safeguards Easily Bypassed by Students

OpenAI's Study mode, designed to guide students through problems using explanations rather than immediately providing answers, proved trivially easy to circumvent. Researchers discovered teenagers could bypass the tutoring approach by selecting a "Show me the answer" option, allowing ChatGPT to complete assignments instead of teaching problem-solving skills1

1

. When testing homework assignments from high school teachers, ChatGPT immediately presented options including working through problems or simply showing answers3

3

. The chatbot also churned out complete essays on literary works without offering to help outline thoughts first3

3

. Study Hours, a feature designed to encourage educational use, also failed to stop workarounds—simply removing the @study prefix from a prompt returned ChatGPT to normal mode1

1

. These safety gaps come amid steep declines in reading and math scores, with recent studies finding that dependence on AI chatbots may stifle students' abilities to retain information and solve complex problems3

3

.

Age Detection System and Privacy Concerns

The age detection system OpenAI created to identify younger users and automatically apply appropriate protections showed significant flaws. Testing found that repeatedly stating "I'm 13" didn't prompt ChatGPT to move accounts from the adult experience to teen safeguards1

1

. A privacy review covering more than 31 OpenAI policy documents revealed that teen conversations are used to train OpenAI's models automatically unless a parent or teen turns that setting off2

2

. Deleting data only stops future training rather than erasing past use, and OpenAI has no privacy policy written specifically for teens2

2

. These privacy concerns add another layer to the ethical considerations surrounding ChatGPT for Teens.

Legal Action and Calls for Digital Citizenship

On October 14, 2024, eight states—Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana—filed coordinated lawsuits against OpenAI in their respective state courts4

4

. The suits allege ChatGPT is an unsafe product that has harmed children and misled parents, from dangerous medical advice to coaching young users towards self-harm4

4

. This represents the most coordinated state action yet against an AI chatbot, coming just one day after a federal class action over children's data4

4

. OpenAI maintains that average teen use is under 15 minutes a day and mostly focused on learning, but critics argue safeguarding must be designed for the teenager who is awake, alone and in trouble—not the average user4

4

. Digital citizenship advocates emphasize that when a child says they're 13 or expresses intent to hurt themselves, those words should be enough to trigger immediate protective responses4

4

. Watch for how OpenAI responds to these findings and whether other AI companies adjust their approaches to teen safety in response to this unacceptable risk rating.