Helping developers build safer AI experiences for teens

OpenAI has released a prompt-based safety policy package to help developers implement protections appropriate for teens. This policy is designed to work with OpenAI's open-weight model gpt-oss-safeguard, enabling developers to easily translate safety requirements into classifiers usable in real systems.

Until now, developers have struggled to convert high-level safety goals into precise operational rules. This is a task that requires both specialized knowledge and AI skills simultaneously, and it can easily lead to protection gaps or inconsistent filtering. This policy was designed to address these issues by reflecting the unique risk factors associated with teens' developmental stages.

The initial release includes policies covering the following key risk categories:

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

This policy was developed with consultation from external expert organizations such as Common Sense Media and everyone.ai. Developers can integrate it into various workflows, including real-time content filtering or offline analysis of user-generated content.