OpenAI Teen Safety Policy Pack and gpt-oss-safeguard

OpenAI has released a set of prompt-based safety policies designed to help developers create age-appropriate protections for teenagers. These policies are built to work with gpt-oss-safeguard-20b, an open-weight safety model, allowing developers to convert high-level safety requirements into usable classifiers for real-world AI systems.

Operationalizing Teen Safety via Prompt-Based Policies

OpenAI provides these safety policies as prompts that can be used directly with reasoning models and the gpt-oss-safeguard model to ensure consistent safety standards. This approach addresses a primary challenge for developers: translating high-level safety goals into precise, operational rules that account for the unique developmental needs and risks associated with teenagers.

By structuring policies as prompts, developers can:

  • Integrate safety checks into existing workflows more easily.
  • Adapt policies to specific use cases.
  • Iterate on safety definitions over time.
  • Use the policies for both real-time content filtering and offline analysis of user-generated content.

Scope of the Initial Safety Policy Release

The initial set of safety policies covers six key risk areas tailored to common challenges faced by teens:

  • Graphic violent content
  • Graphic sexual content
  • Harmful body ideals and behaviors
  • Dangerous activities and challenges
  • Romantic or violent roleplay
  • Age-restricted goods and services

Expert Collaboration and Development

These policies were developed with input from external organizations, including Common Sense Media and everyone.ai, to refine the scope of content, strengthen prompt structures, and identify critical edge cases.

Regarding the impact of these tools, Robbie Torney, Head of AI & Digital Assessments at Common Sense Media, stated:

"One of the biggest gaps in AI safety for teens has been the lack of clear, operational policies that developers can build from. Many times, developers are starting from scratch. These prompt-based policies help set a meaningful safety floor across the ecosystem..."

Dr. Mathilde Cerioli, Chief Scientist at everyone.AI, noted that these policies help "translate expert knowledge into guidance that can be used in real systems."

Implementation as a "Defense in Depth" Strategy

OpenAI emphasizes that these policies are a starting point rather than a complete safety solution. Because every application has unique risks and contexts, developers are encouraged to adopt a "defense in depth" approach, combining these prompt-based policies with additional safeguards, including:

  • Product design decisions
  • User controls
  • Teen-friendly transparency
  • Monitoring systems
  • Age-appropriate responses

These released policies are based on OpenAI's internal experience but do not represent the full extent of the company's internal safeguards.

Sources