OpenAI Launches ChatGPT for Teens: Default Safeguards Still Lack Outcome Data

If the default safeguards in ChatGPT for Teens work as intended, the share of harmful interactions involving self-harm advice, emotional-dependence cues, or inappropriate sexual content should decline over the next three to six months. OpenAI has not published a pre-launch incident baseline, interception rate, or false-positive rate, so that hypothesis cannot yet be tested.

OpenAI launched ChatGPT for Teens on August 18, 2026 for users aged 13 to 17. The company says the system restricts conversations involving suicide, self-harm, romance, and sexual content. It also prevents the chatbot from suggesting that it has personal feelings, consciousness, or emotions. Homework and study support is designed to guide students toward their own answers instead of producing essays for them. These changes alter the default behavior of teen accounts; a parent does not have to activate controls before the core restrictions apply.

Age assurance determines who enters the teen experience

OpenAI does not verify every user’s age. The system estimates whether someone is under 18 from signals that include the types of queries they submit. A user who identifies as a minor, or whom the system classifies as one, is automatically placed into the teen experience. That mechanism can miss teenagers who are not detected and can route adults incorrectly. Neither the company announcement nor the Associated Press report provides the age classifier’s accuracy, the time required to resolve an appeal, or coverage by market.

When both the teenager and a parent or guardian agree to link their accounts, parents can set quiet hours during which ChatGPT is unavailable. They can also receive safety notifications in a limited set of high-risk situations, including possible self-harm. OpenAI says it is adding notifications related to eating disorders while limiting the information shared. These tools require both sides to opt in. Brendan Bouffard, an associate general counsel at the children’s advocacy group Fairplay, told AP that some protections he considers important, including restrictions on ChatGPT’s long-term memory, are not enabled by default and still require parental intervention.

A feature list does not establish lower risk

AP cited a 2025 Common Sense Media study reporting that more than 70% of U.S. teenagers had used AI chatbots for companionship and that half used AI companions regularly. That level of use makes limits on emotional-dependence cues commercially and socially relevant, while also increasing the cost of classification errors. OpenAI has described the content and model behaviors it intends to block, but it has not released red-team prompts, failure cases by age, or results from an independent audit.

The product also sits between two competing requirements. A teen account must offer enough utility for study and ordinary questions to retain users, yet it must intervene when a conversation becomes dangerous without turning every sensitive question into an alarm. Excessive blocking could push teenagers toward adult accounts or other services. Weak blocking would leave the harms that motivated the dedicated product largely unchanged. The current sources do not quantify either failure mode.

Over the next three to six months, four measures would make the launch assessable: harmful-response rates for teen accounts, age-classification error rates, the share of safety notifications that lead to offline support, and the share of study sessions in which students complete the task themselves. If OpenAI continues to publish only feature descriptions, families and schools will still be unable to translate “stronger protections” into a comparable reduction in risk.

Sources:

• OpenAI: Introducing ChatGPT for Teens: Built for learning, backed by protections
• Associated Press: OpenAI launches ChatGPT for Teens, with content restrictions and study help