OpenAI Model Spec Update: Under-18 (U18) Principles

OpenAI has updated its Model Spec—the set of rules and behavioral expectations guiding AI model behavior—to include Under-18 (U18) Principles. These principles ensure that ChatGPT provides a safe, age-appropriate experience for users aged 13 to 17 by prioritizing prevention, transparency, and early intervention grounded in developmental science.

Core Commitments of the U18 Principles

The U18 Principles clarify how the Model Spec applies to teen contexts where safety considerations for minors are more pronounced. The framework is anchored by four guiding commitments:

  • Prioritizing Teen Safety: Safety is placed first, even when it conflicts with other model goals.
  • Promoting Real-World Support: The model is designed to encourage offline relationships and the use of trusted resources.
  • Age-Appropriate Interaction: The model treats teens as teens, avoiding both condescension and treating them as adults.
  • Transparency: The system sets clear expectations for the user.

Safety Guardrails and High-Risk Areas

Consistent with the Teen Safety Blueprint, OpenAI has implemented safeguards to ensure the model takes extra care when discussing high-risk topics. These protections apply to users who identify as under 18 at sign-up and are managed through parental controls.

Specific high-risk areas include:

  • Self-harm and suicide
  • Romantic or sexualized roleplay
  • Graphic or explicit content
  • Dangerous activities and substances
  • Body image and disordered eating
  • Requests to keep secrets about unsafe behavior

When safety concerns arise, teens encounter stronger guardrails, safer alternatives, and encouragement to seek offline support. In cases of imminent risk, the model urges users to contact emergency services or crisis resources.

Expert Collaboration and Clinical Guidance

OpenAI developed these principles by previewing them with external experts, including the American Psychological Association (APA). Dr. Arthur C. Evans Jr, CEO of the APA, emphasized that AI tools should be balanced with human interactions critical for social and psychological development:

"APA encourages AI developers to offer developmentally appropriate precautions for youth users of their products and to take a more protective approach for younger users. Children and adolescents might benefit from AI tools if they are balanced with human interactions that science shows are critical for social, psychological, behavioral, and even biological development."

Additional guidance is provided by the Expert Council on Well-Being and AI, which advises on healthy AI interactions across all ages, and the Global Physician Network, which helps the model better recognize distress and guide users toward professional care.

Product Integration and Future Roadmap

Teen protections have been extended across multiple products, including group chats, the ChatGPT Atlas browser, and the Sora app. To further automate these protections, OpenAI is rolling out an age prediction model on ChatGPT consumer plans. This model will automatically apply teen safeguards when an account is believed to belong to a minor; if age cannot be confidently verified, the system will default to the U18 experience.

These safety measures build upon foundations established in GPT-5.2 and include partnerships with ThroughLine to surface localized helplines within ChatGPT and Sora.

Sources