OpenAI is rolling out a new set of prompt-based safety policies aimed at helping developers build more age-appropriate AI experiences for teenagers. These policies are engineered to integrate with the company’s open-weight safety model, gpt-oss-safeguard, simplifying the process of turning complex safety requirements into actionable classifiers for real-world applications.
The move underscores OpenAI's commitment to balancing innovation with responsible deployment, particularly for younger users. The company believes that providing developers with capable models and robust safety tools is crucial for fostering a safer AI ecosystem. These new policies are a direct extension of OpenAI's broader efforts, including updates to its Model Spec with Under-18 principles and product-level safeguards like parental controls.
Translating Safety Needs into Actionable Prompts
A primary challenge in AI safety is translating broad policy goals into concrete, enforceable rules. Developers often struggle to define precisely what constitutes teen-specific risks, leading to potential gaps or over-filtering. OpenAI’s new approach addresses this by structuring safety policies as prompts, which can be directly fed into models like gpt-oss-safeguard.
This format enables developers to more easily implement consistent safety standards across their systems. The initial release covers critical areas such as graphic violent and sexual content, harmful body ideals, dangerous challenges, romantic or violent roleplay, and age-restricted goods and services. These policies can be utilized for both real-time content moderation and offline analysis.
Expert Input Shapes Teen Safeguards
OpenAI collaborated with external organizations, including Common Sense Media and everyone.ai, to develop these policies. Their expertise was instrumental in defining the scope, refining prompt structures, and identifying edge cases. This collaborative approach aims to establish a meaningful safety baseline across the AI development community.