Underage sexual content, self-harm info targeted by OpenAI’s new open-source prompts

OpenAI has announced new open source security prompts for developers, aiming to kick off a massive rollout of policies aimed at protecting teens.

The prompt-based safety pack includes model advice on common teen risks, developmental content recommendations, and age-appropriate guidelines on topics such as self-harm, sexual content and romantic role-play, dangerous trends or viral challenges, and harmful body ideals.

OpenAI said this is a more robust alternative to previously proposed high-level guidelines, formatted as prompts that connect directly to AI systems.

OpenAI added new under-18 principles to its model specification in December. A few months earlier, the company released gpt-oss-safeguard, an open reasoning model designed to help developers implement security conditions and classify safe and unsafe content. Unlike traditional security classification processes, gpt-oss-safeguard can be fed directly by platform security policies and infers intent by distinguishing appropriate outcomes.

But “even experienced teams often struggle to translate high-level security objectives into precise operational rules, especially since this requires both subject matter expertise and deep knowledge of AI,” OpenAI said in its latest press release. “This can lead to gaps in protection, inconsistent enforcement, or overbroad filtering. Clear, well-targeted policies are an essential foundation for effective security systems.”

The additional development pack was designed in collaboration with Common Sense Media, a non-profit organization, and Everyone.ai.

Experts have warned parents against over-exposing vulnerable teenagers and even young children to chatbots, as AI companies try to get a handle on the ramifications of their models on users’ mental health. Last year, OpenAI was sued by the parents of teenager Adam Raine in the industry’s first wrongful death case, with the Raine family claiming that a combination of ChatGPT sycophancy and lax security policies were responsible for their son’s suicide death. The company denied allegations of wrongdoing and, in response, strengthened its teen mental health and safety features, including age assurance. Despite this, third-party developers licensing OpenAI models have struggled to maintain the same level of safety precautions, including in AI-powered children’s toys.

The case against OpenAI follows several lawsuits against the controversial Character.AI platform and set the stage for a recent wrongful death lawsuit filed against OpenAI competitor Google and its Gemini AI assistant.

Technology and social media companies industry-wide are facing a wave of legal challenges over the long-term impact of their products on users. Last month, Instagram CEO Adam Mosseri and Meta chief Mark Zuckerberg testified before a jury in a landmark case putting social media platforms on trial for their allegedly addictive design principles. A verdict has not yet been rendered.

OpenAI said its new safety prompt pack is not a “complete or definitive definition or guarantee of teen safety.” Robbie Torney, head of AI and digital assessments at Common Sense Media, said the new policies can create a “meaningful security floor across the ecosystem,” filling an AI security gap that has been exacerbated by the lack of operational policies for developers.

Developers can download OpenAI’s security template on Hugging Face and access its new prompt pack on GitHub.

Disclosure: Ziff Davis, Mashable’s parent company, filed a lawsuit in April against OpenAI, alleging that it violated Ziff Davis’ copyrights in the training and operation of its AI systems.