OpenAI's New Under-18 Principles Enhance AI Ethics and Teen Safety in ChatGPT

On December 18, 2025, OpenAI updated its Model Spec—the written set of behavioral expectations that guides how ChatGPT should respond—by adding a new section: Under-18 (U18) Principles. The goal is straightforward: teens (ages 13–17) have different developmental needs than adults, and a “one-size-fits-all” safety posture can create gaps in higher-risk situations.

At a high level, the update clarifies how existing safety rules apply in teen conversations and adds age-appropriate guidance where needed. The principles emphasize prevention, clearer boundaries, and stronger encouragement toward real-world support when risks show up.

This article explains what the U18 Principles are, why they matter, and what “safe, age-appropriate behavior” looks like in practice—without turning teen safety into vague slogans. If you’re interested in related context on teen safety work, you may also want to read: OpenAI’s Teen Safety Blueprint.

TL;DR

• What changed: OpenAI added Under-18 (U18) Principles to the Model Spec to guide ChatGPT’s behavior for teen users (13–17).
• What they emphasize: teen safety first, real-world support, age-appropriate tone, and transparency about boundaries.
• Where it matters most: higher-risk topics (e.g., self-harm, sexualized roleplay, dangerous activities/substances, disordered eating/body image, and requests to keep unsafe secrets).

What the Model Spec Is (and Why Updating It Matters)

Model behavior isn’t only a “model capability” issue; it’s a product safety issue. A model can be powerful and still be unsafe in certain contexts if the rules for boundaries, refusals, and support are unclear. The Model Spec acts as a written reference describing how models should behave—especially in difficult or high-stakes situations. This update introduces teen-specific guidance inside that same framework, rather than treating teen safety as an informal add-on.

Practically, written principles matter because they:

• create consistency across products and teams
• make behavior expectations auditable and testable
• reduce “policy drift” when systems evolve
• help users understand what to expect and what not to expect

Why Teens Need a Different Safety Lens

Teen users often arrive with different needs than adults: they’re learning boundaries, building identity, and navigating social and emotional challenges. The U18 Principles aim to deliver support that is neither condescending nor adult-coded, while applying stronger protections when risk signals appear.

In practice, teen safety isn’t only about “blocking bad content.” It’s about guiding conversations in a way that:

• reduces exposure to high-risk content and harmful spirals
• encourages healthier offline support systems
• keeps boundaries clear, especially when the user requests secrecy or risky escalation
• does not treat the AI as a replacement for real-world help

For educators thinking about safe classroom use, this related piece may be helpful: Exploring ChatGPT for Teachers (Secure Use).

The Four Core Commitments in the U18 Principles

The U18 Principles can be understood as four guiding commitments. They provide a simple mental model for what “age-appropriate” means beyond a list of forbidden topics.

1) Put teen safety first

Teen safety is prioritized even when it conflicts with other goals. In practical terms, this means stronger caution in higher-risk contexts and less tolerance for interactions that could intensify harm.

2) Promote real-world support

When credible risks arise, the principles emphasize encouraging offline relationships and trusted resources. The model can provide general guidance, but it should not position itself as a substitute for real-world care.

3) Treat teens like teens

This commitment aims for balance: respectful and clear communication without speaking down to teens—and without treating them as adults when the context calls for stronger protections.

4) Be transparent with boundaries

Transparency here means setting clear expectations: what the assistant can help with, what it will refuse, and why. Clear boundaries reduce confusion and can discourage “workarounds” in risky situations.

Which Topics Get Stronger Guardrails

The update highlights higher-risk areas where the system should take extra care with teen users. The goal is not to panic; it’s to recognize where harm is more likely and to respond with prevention and early intervention.

Examples of areas often treated as higher-risk include:

• self-harm and suicide-related conversations (handled with extra caution and emphasis on real-world support)
• romantic or sexualized roleplay and requests for explicit content
• dangerous activities and substance misuse
• body image and disordered eating contexts
• requests to keep secrets about unsafe behavior

If you’re building agentic or automation-heavy systems, the “clear boundary + refusal when risky” approach here aligns with broader responsible automation thinking: Setting Boundaries for Automation in Productivity.

Why “Transparency” Is a Safety Feature (Not PR)

When users (especially teens) don’t understand a system’s limits, they can interpret helpful tone as authority or emotional certainty. The U18 Principles emphasize clarity: expectations, boundaries, and encouragement toward real-world support when it matters.

In practice, transparency looks like:

• saying what the system can do (general guidance, coping strategies, learning support)
• saying what it cannot do (replace professional care, keep unsafe secrets, participate in harmful roleplay)
• staying calm and direct rather than dramatic or overly validating in risky contexts

What This Means for Parents and Guardians

For families, the key value is clarity: a stronger expectation that teen safety and age-appropriate responses come first.

Practical ways to use this update as a parent/guardian:

• Talk about boundaries: explain that AI can help brainstorm, learn, and organize—but it’s not a private therapist or a replacement for trusted adults.
• Normalize asking for help: especially for situations involving safety, anxiety spirals, or risky pressure.
• Encourage disclosure culture: teens should feel safe telling a trusted person when something online feels disturbing or manipulative.

What This Means for Schools and Youth Programs

In education contexts, teen-specific safety principles reinforce an important message: AI is not just a “learning tool,” it’s a socio-technical system with real responsibility around privacy, safety, and appropriate guidance.

School-friendly actions that align with the U18 approach:

• define what AI is allowed for (study plans, practice questions, drafting outlines) vs not allowed for (submitting unedited work, sensitive counseling)
• avoid placing personal student details into general AI tools
• teach “AI literacy” as a safety skill: verifying claims, recognizing manipulation, and asking for human support

If you’re building or refining an ethics posture for large-scale tools, this internal article can complement the U18 topic: Ethical Challenges and Considerations in Real Deployments.

What This Means for Product Teams and Developers

Serving minors changes product requirements. It’s not enough to “have a model.” You need guardrails that are testable, enforceable, and continuously improved.

A practical developer checklist aligned with the U18 spirit:

• Define teen-safe defaults: when age is known or strongly implied, behave more cautiously by default.
• Strengthen refusal logic: especially for high-risk categories (danger, explicit sexual content, unsafe secrecy).
• Prefer prevention over debate: reduce “argue with the system” dynamics in risky contexts; keep responses supportive and bounded.
• Instrument safety: log patterns, run evaluations, and update mitigations when failures are discovered.
• Make escalation easy: design UX that encourages real-world support when risk signals appear.

FAQ

What are the Under-18 (U18) Principles?

They are guidelines added to OpenAI’s Model Spec describing how ChatGPT should provide a safer, age-appropriate experience for teen users aged 13–17, informed by developmental considerations.

Do the regular Model Spec rules still apply?

Yes. The U18 Principles build on existing safety rules and clarify how those rules should be applied in teen conversations, especially in higher-risk situations.

What’s the biggest practical change for teen conversations?

Clearer boundaries, stronger safety-first behavior, and more emphasis on encouraging real-world support when credible risks arise—plus stricter handling of certain higher-risk topics.

Summary

OpenAI’s December 18, 2025 update adds Under-18 (U18) Principles to the Model Spec to guide safer, age-appropriate support for teens. The update can be summarized as four commitments—teen safety first, real-world support, treating teens like teens, and transparency—and it highlights higher-risk areas where guardrails should be stronger.

External reference: Updating our Model Spec with teen protections and model-spec.openai.com.