OpenAI introduced ChatGPT for Teens with stronger restrictions around suicide, self-harm, romantic and sexual conversations and with homework features intended to support learning rather than simply supply answers. OpenAI introduced a teen-focused experience with tighter limits on self-harm, sexual and romantic conversations and a study-oriented design. The teen-focused version was introduced during the week of August 18.
OpenAI said it applies stronger restrictions to suicide and self-harm discussions. The company also described tighter controls on romantic and sexual conversations. The product announcement establishes the intended rules and interface. It cannot establish the frequency of successful refusals, false reassurance or workaround behavior, which require systematic testing across languages, contexts and repeated conversations.
Study and homework behavior is designed to emphasize explanation and learning rather than direct output alone. The product arrives as the first large cohort of students grows up with widely available generative AI. The announced safeguards are company claims and do not independently prove performance in every conversation. Educational design also needs separate evaluation from crisis safety. A system might guide homework effectively while mishandling emotional reliance, or block risky dialogue while producing inaccurate academic explanations. Evidence should report those dimensions independently.
Age assurance, parental controls, model behavior and escalation to human help are separate components of youth safety. A refusal or warning can reduce one risk while still leaving questions about accuracy, privacy and emotional dependency.
Independent testing must use age-appropriate methods and avoid exposing minors to unnecessary harmful content. The current evidentiary limit is that real-world safeguard performance, age-verification reliability and comparative learning outcomes were not yet independently established.
The next factual record will come from transparent safety evaluations and incident reporting and school, parent and youth experience with the teen-specific controls. Until those records appear, the account remains bounded by the cited reporting, measurements and explicitly attributed statements.