OpenAI announced on Tuesday the launch of a youth-oriented safety policy prompt toolkit for developers, aiming to help third-party applications incorporate safety rules to protect minors from the beginning of their design. The toolkit is available as a set of directly callable "Youth Safety Policy" prompts and can be used with OpenAI's open source weighted safety model gpt-oss-safeguard.
OpenAI said that developers do not need to start from scratch on how to design safety mechanisms for minors, and can directly use this set of prompt words to "reinforce" existing or newly developed AI applications. These policies focus on sensitive areas such as graphic violence and pornographic content, harmful body image and behavior, risky activities and challenges, romantic or violent role play, and age-restricted goods and services. The company emphasizes that these security policies are provided in the form of prompt words, so they can be easily adapted to other models other than gpt-oss-safeguard, but the effect is expected to be better when used within OpenAI's own ecosystem.
In developing the policy, OpenAI works with third parties in the youth and content safety space, including AI safety watchdog Common Sense Media and everyone.ai. Robbie Torney, head of AI and digital assessment at Common Sense Media, said in a statement that these prompt word-based policies help establish a meaningful security bottom line for the entire ecosystem, and the open source release method also allows all parties to continuously adjust and improve over time.
OpenAI points out in a companion blog that even experienced teams often struggle in practice to translate abstract security goals into concrete, enforceable rules. The company believes that this deviation can lead to loopholes in protection measures, inconsistent implementation standards, and even excessive blocking, so clear and well-bounded policies are a key foundation for building an effective security system.
OpenAI also acknowledges that this set of policies cannot fundamentally solve all the complex challenges of AI safety. But the company emphasized that the new tool is a further expansion of its existing security measures, including product-level parental controls, age prediction and other functions. In 2025, OpenAI updated the behavior specification (Model Spec) of its large language model to clarify how the model should respond and limit when facing users under the age of 18.
Still, OpenAI faces questions about its own safety record. The company is currently facing multiple lawsuits filed by family members of users who committed suicide after extreme use of ChatGPT. The lawsuit alleges that some of the dangerous relationships formed after users bypassed chatbot security, and that no current model's defenses are truly "impenetrable." Against this background, the industry generally believes that although this open source youth safety policy prompt is not the ultimate solution, it provides a relatively easy-to-use compliance path for independent developers and small teams, and is regarded as a new step forward in promoting AI safety practices for minors.
learn more:
https://github.com/openai/teen-safety-policy-pack