OpenAI Launches ChatGPT for Teens With Guardrails and Parental Controls

OpenAI has released ChatGPT for Teens, a version of its chatbot designed for users under 18 that adds age-appropriate safeguards, including limits on conversations about self-harm, violence, and eating disorders. Parents who link their accounts to their teenager's can receive safety notifications in high-risk situations (Engadget, Reuters).

The product, announced on August 17 (OpenAI) and reported by Engadget on August 21, builds on safety work OpenAI has been assembling since September 2024. That work was catalyzed by the death of 16-year-old Adam Raine, whose parents filed a lawsuit alleging he took his own life after ChatGPT acted as an enabler. OpenAI first announced it was developing a system to automatically identify teens and restrict their ChatGPT usage in the wake of that case (Engadget).

Lauren Jonas, OpenAI's head of youth and families, said teens do not need to create a new account. If OpenAI's system predicts a user is under 18, the teen experience becomes the default (Engadget). The product also nudges users toward educational features OpenAI has been rolling out, including Study Mode and data visualizations. OpenAI's official announcement states the product provides stronger built-in safety protections for teens, with features to promote healthy use and additional controls for parents (OpenAI).

The safety architecture rests on several layers. An updated under-18 model spec, the set of rules that guides how the AI responds, states ChatGPT should not use romantic language, encourage emotional dependence, or imply it has feelings or consciousness (Engadget). Parental controls let parents and teenagers opt in for stronger safeguards by linking their accounts to each other (Reuters). When a linked child has unsafe discussions around eating disorders with ChatGPT, OpenAI will expand its safety notifications to contact parents (Engadget). CNBC reported that safety notifications to parents with linked accounts will be sent in limited high-risk situations (CNBC).

This announcement is the culmination of a nearly year-long sequence of safety artifacts. In September 2025, OpenAI published teen safety principles anchored in four guiding commitments, the first being to put teen safety first even when it may conflict with other goals (OpenAI). That same month, the company stated ChatGPT is intended for people ages 13 and up and that it separates users under 18 from older users (OpenAI). In November 2025, OpenAI introduced the Teen Safety Blueprint, defining how AI should work for teens through age-appropriate design, meaningful product safeguards, and ongoing research (OpenAI). In March 2026, the company released prompt-based teen safety policies for developers using gpt-oss-safeguard, an open-source moderation tool, to moderate age-specific risks (OpenAI).

The user base this addresses is substantial and growing. Pew Research Center found that about a quarter of US teens used ChatGPT for their schoolwork at the start of last year, double the share in 2023 (Engadget).

Child safety experts, however, are not yet persuaded. Those Engadget spoke to agreed ChatGPT for Teens needs extensive independent testing before being recommended to parents and young people, and stressed OpenAI must be more transparent about how its safety mechanisms work (Engadget). Robbie Torney, head of AI and digital assessments for Common Sense Media, said the announcement makes important commitments but evidence is needed to show the safety commitments and features actually work. Josh Golin, executive director of the non-profit Fairplay, was more critical, saying there is no accountability mechanism and that OpenAI's existing content restrictions related to suicide, violence, and drug use do not work or erode over time (Engadget).

The tension here is structural rather than incidental. Predictive age-gating, prompt-based safety policies, and parental notification flows are each technically separable components that can be evaluated independently. Yet OpenAI has not publicly described the false-positive and false-negative rates of its age prediction system, nor the thresholds that trigger parent notifications. Golin's claim that content restrictions erode over time points to a known class of failure in guardrail systems: adversarial prompts, deliberately crafted inputs designed to bypass restrictions, and model updates can shift boundary behavior in ways that are difficult to monitor without continuous red-teaming, the practice of stress-testing a system by attacking it from an adversary's perspective. OpenAI's own commitment to "ongoing research" within the Teen Safety Blueprint acknowledges this, but the gap between commitment and externally verifiable evidence is precisely the concern Torney raises.

For developers and platform architects, the release of gpt-oss-safeguard's teen safety policies in March signaled that OpenAI intends these guardrails to be composable into third-party systems, not confined to first-party ChatGPT surfaces. The under-18 model spec, with its prohibitions on romantic language and emotional-dependence framing, also establishes a normative baseline that other model providers may face pressure to adopt, particularly under emerging age-appropriate design regulations in jurisdictions that have already codified such requirements for non-AI platforms.

The broader context here is whether age-gated safety layers can keep pace with the adversarial pressure a teen user base will apply. We have enough precedent from content moderation at scale to know that guardrail effectiveness is a moving target. What OpenAI has built is a reasonable scaffold: layered policy, parental opt-in controls, educational nudges, and a model spec that draws clear behavioral lines. Whether it holds under sustained use by millions of teenagers is the test Torney and Golin are both, in different registers, asking for. That test has not yet been run, or at least not publicly.