OpenAI puts teen safety 'first' in new ChatGPT model specification

Updated Guidelines Prioritize Minor Protection Over General Usefulness, Introduces AI-Based Age Prediction System

OpenAI has announced a major update to its internal rules, or Model Spec, establishing explicit U18 Principles that mandate ChatGPT's behavior must prioritize the safety of users aged 13 to 17 above all other objectives, including the platform's traditional goal of maximizing helpfulness and freedom.

The move comes amid mounting public and regulatory pressure, including recent lawsuits, regarding the psychological impact of generative AI on vulnerable young minds.

The core change: Safety over helpfulness

The updated Model Spec creates a specific set of guidelines for teenagers (13-17), guided by developmental science and input from experts like the American Psychological Association (APA). The key principles guiding this shift are:

• Safety first mandate: ChatGPT must "Put teen safety first, even when it may conflict with other goals." This is the most significant change, ensuring the AI will be more cautious and restrictive in high-risk conversations.

• Promote real-world support: The model is now instructed to actively encourage teens to seek help from trusted adults (parents, teachers) or professional crisis resources, rather than positioning itself as a substitute for human relationships or therapy.

• Age-appropriate Interaction: The AI must avoid being condescending while simultaneously avoiding treating a teenager like a mature adult, especially in sensitive discussions.

• Transparency: ChatGPT is required to be clear about its limitations and what it can and cannot do for a young user.

Stricter guardrails in sensitive areas

Under the new rules, ChatGPT will apply stronger safeguards when discussing topics deemed higher-risk for teens:

• Self-harm and suicide: Stronger guardrails and immediate redirection to crisis hotlines.

• Explicit content: Increased filtering of graphic, sexual, or violent content, and blocking of romantic or sexualized roleplay.

• Body image: Restrictions on discussions promoting extreme beauty standards, unhealthy dieting, or disordered eating.

• Secrets: The AI will not agree to keep secrets that involve potentially harmful behavior.

• Automatic age detection underway

To ensure these protections are applied automatically, OpenAI is in the early stages of rolling out an Age Prediction Model.

This AI-based system analyzes subtle linguistic cues and conversational patterns to estimate whether a user is under 18, even if they haven't explicitly disclosed their age. If the system is not confident about a user's age, it will default to the safer, U18 experience.

Adults who are mistakenly flagged will be given an option to verify their age to restore full access.

The new rules and tools build on previous steps, including the introduction of parental controls that allow parents to link and customize their teen's ChatGPT experience.