OpenAI releases teen-safety policy prompts for gpt-oss-safeguard

What happened

OpenAI said on March 24, 2026 that it is releasing a set of prompt-based safety policies to help developers build age-appropriate AI experiences for teens. The package is designed to work directly with the company’s open-weight safety model, gpt-oss-safeguard, and OpenAI presents it as a practical bridge between high-level safety goals and deployable classifiers that can be used in real systems.

The announcement is notable because it turns youth-safety guidance into something operational. Instead of only publishing broad principles, OpenAI is shipping policy prompts that developers can use for real-time filtering or offline analysis of user-generated content. That makes the release more relevant to product teams that need implementation details, not just policy language, especially as open-weight models spread into education, social apps, creator tools, and consumer assistants.

What is included

OpenAI says the first policy set covers six areas: graphic violent content, graphic sexual content, harmful body ideals and behaviors, dangerous activities and challenges, romantic or violent roleplay, and age-restricted goods and services. The company also says the material can be adapted for other reasoning models, which means the work is positioned less as a one-off feature and more as reusable safety infrastructure.

Another important part of the release is the ecosystem framing. OpenAI says the policies were developed with input from Common Sense Media and everyone.ai, and that they are being released as open source through the ROOST Model Community. That matters because youth safety has often been discussed as a product responsibility held by a few large platforms. By pushing policy prompts into the open, OpenAI is effectively arguing that baseline teen-protection tooling should be shared across the broader open-model stack.

What to watch

OpenAI is explicit that these prompts are a starting point rather than a complete solution. The company says developers still need layered safeguards, product decisions, monitoring, and age-appropriate controls around the model itself. The practical question now is whether other builders adopt the prompts, translate them, and extend them to more categories, which would turn this release from a documentation update into a common safety layer for youth-facing AI products.