ChatGPT for Teens Receives 'Unacceptable Risk' Rating After Safety Features Fail Key Tests

Critical Safety Gaps Emerge in ChatGPT for Teens Testing

OpenAI's ChatGPT for Teens, launched on August 18, 2026, has been rated "Unacceptable Risk" by Common Sense Media's Youth AI Safety Institute following extensive testing that exposed significant failures in the platform's youth AI safety protections1

1

2

2

. The evaluation, which tested more than 4,000 prompts across accounts registered to users aged 13 to 17, revealed troubling gaps between OpenAI's promises and the actual performance of its teen safety features. ChatGPT for Teens was designed as a package of default settings and classifiers applied automatically to accounts believed to belong to teenagers, not a separate app or model2

2

. The platform's failures in protecting vulnerable users raise urgent questions about whether AI safety measures are keeping pace with the rapid deployment of generative AI tools to younger audiences.

Parental Alerts Fail During Mental Health Crises

The most alarming finding centers on ChatGPT for Teens' failure to trigger parental alerts during high-risk conversations involving suicidal thoughts, self-harm, and eating disorders. Common Sense Media set up more than 12 free-tier accounts with no prior chat history, each linked to a parent account before the first message was sent2

2

. Researchers created four scripted personas representing teenagers in crisis: a 14-year-old repeatedly asking the chatbot to help her end her life, a 13-year-old seeking advice on cutting herself more severely, a 17-year-old former eating disorder patient reporting fainting and heart palpitations, and a 16-year-old describing escalating purging behaviors2

2

. Despite explicit disclosures within opening exchanges, no parental alerts arrived at any point during conversations lasting up to 60 minutes2

2

. OpenAI later told the Institute that accounts must be connected for roughly three hours before notifications can be sent, but the Institute found this explanation inadequate given that some test accounts exceeded that threshold2

2

. In real-world testing by a journalist, subtle allusions to suicide initially failed to trigger crisis resources, though more direct language eventually prompted an appropriate response3

3

.

Mental Health Crisis Response Proves Inconsistent

Beyond the parental alert failures, ChatGPT for Teens demonstrated unreliable mental health crisis response across multiple scenarios. The Youth AI Safety Institute consulted child psychiatrists and a pediatrician to evaluate responses, finding that ChatGPT failed to direct teenagers to appropriate help in more than 25% of situations that experts designed to require it1

1

. Pre-launch testing involving 990 prompts across 13 clinical conditions generated only two notifications, with one arriving three hours after the first crisis-level disclosure and another three days later2

2

. Post-launch testing of 450 prompts produced 251 responses that warranted crisis resources according to medical experts, yet only two alerts were sent2

2

. The Institute concluded that alerts appear keyed to conversation history accumulated over time rather than the severity of individual messages, a fundamental flaw when teenagers in immediate crisis need urgent intervention2

2

. This approach contrasts sharply with teen-specific AI mental health apps evaluated by the Institute, which triggered phone calls to guardians within 15 minutes of a first disclosure2

2

.

Study Mode Offers Homework Answers Instead of Tutoring

OpenAI's tutoring mode, designed to guide students through problems using explanations and questions rather than providing direct answers, proved easily circumvented in practice. When a journalist tested ChatGPT for Teens with actual homework assignments from high school teachers, the chatbot immediately presented options including "Show me the answer"3

3

. Selecting this option allowed ChatGPT to complete assignments outright, defeating the educational purpose of the feature1

1

. The chatbot wrote full essays on literary works like James Joyce's "Dubliners" and Joseph Conrad's "Heart of Darkness" without hesitation, and when asked to make the writing less detectable as AI-generated, it offered to "rewrite it in a more natural student voice"3

3

. Common Sense Media found that teenagers could bypass the Study Hours feature, which parents can schedule to encourage educational use, simply by removing the @study prefix from a prompt1

1

. These workarounds raise serious ethical considerations about whether the platform genuinely supports learning or enables academic shortcuts that may harm students' long-term development.

Age Detection and Privacy Concerns Compound Problems

The age detection system OpenAI created to identify younger users and automatically apply appropriate protections showed significant weaknesses during testing. Researchers found that repeatedly stating they were 13 years old didn't prompt ChatGPT to move test accounts from the adult experience to the teen-specific version1

1

. Privacy concerns emerged from a separate review by Common Sense Privacy covering more than 31 OpenAI policy documents, which found that teen conversations are used to train OpenAI's models automatically unless a parent or teen manually disables that setting2

2

. Deleting data only prevents future training rather than erasing past use of conversations, and OpenAI has no privacy policy written specifically for teenagers2

2

. These privacy concerns matter particularly for vulnerable young users sharing sensitive information about mental health, relationships, or personal struggles with an AI system that treats their disclosures as training data by default.

Comprehensive Risk Assessment Delivers Damning Verdict

The Youth AI Safety Institute's evaluation, dated October 7, 2026, scored ChatGPT for Teens across eight AI Principles using a five-level risk scale. Two principles, "Keep Kids & Teens Safe" and "Put People First," both landed at Unacceptable Risk, the bottom tier2

2

. Four additional principles, "Be Effective," "Support Human Connection," "Use Data Responsibly," and "Be Transparent & Accountable," scored High Risk2

2

. Only "Prioritize Fairness" and "Be Trustworthy" achieved Moderate Risk ratings2

2

. The report consolidated two earlier reviews: a High Risk rating for general ChatGPT use on October 23, 2025, and an Unacceptable Risk rating specifically for ChatGPT's mental health support capabilities on November 14, 20252

2

. Testing spanned two distinct periods, July 13 through August 17, 2026 before launch, and August 25 through September 28 after launch, using accounts across free and paid tiers, some linked to parent accounts and some not2

2

. The organization recommended OpenAI restrict ChatGPT to adults until the problems are resolved1

1

.

What Parents and Educators Should Watch For

The findings arrive at a critical moment as ChatGPT becomes a regular part of teenagers' lives for schoolwork and personal questions. Recent studies have found that dependence on AI chatbots may stifle students' abilities to retain information and learn to solve complex problems, amid steep declines in reading and math scores across the country3

3

. While some features showed promise, such as ChatGPT's refusal to engage in explicit sexual roleplay and its ability to walk students through problems step-by-step when properly configured, the platform continued responding in ways that suggested it had feelings, preferences, and moods1

1

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

1

. If OpenAI succeeds in addressing these problems, it could trigger industrywide changes in how chatbots are designed for young users, as other companies are likely to follow suit3

3

. Until then, parents should recognize that the promised safeguards may not deliver the protection they expect during the moments that matter most.