OpenAI launches open-source tools for AI developers focused on teen safety.

      OpenAI has been dealing with lawsuits over the past year from families of young individuals who died following prolonged interactions with ChatGPT. The company is now taking steps to provide developers with tools to prevent similar issues from arising.

      On Tuesday, OpenAI announced the release of open-source, prompt-based safety policies aimed at helping developers create safer AI applications for teenagers. These policies are intended for use with gpt-oss-safeguard, OpenAI’s open-weight safety model, but they are designed as prompts compatible with other models as well.

      What the policies address

      The prompts focus on five types of harm that AI systems can impose on younger users: graphic violence and sexual content, harmful body images and behaviors, dangerous activities and challenges, romantic or violent role-playing, and access to age-restricted goods and services. Developers can implement these policies into their systems instead of having to create safety rules for teens from the ground up, a task that OpenAI recognized is often mishandled even by experienced teams.

      OpenAI created these policies in partnership with Common Sense Media, a prominent child safety advocacy group, and everyone.ai, an AI safety consulting firm. According to Robbie Torney, head of AI and digital assessments at Common Sense Media, the prompt-based strategy aims to establish a foundational standard across the developer ecosystem, allowing for adaptation and enhancement over time since the policies are open-source.

      The EU tech scene’s latest happenings include insights from our founder Boris and some questionable AI-generated art. This weekly newsletter is free and filled with updates—sign up today!

      OpenAI acknowledged the practical challenges developers face. In a blog post accompanying the announcement, the company noted that developers frequently find it difficult to convert safety objectives into specific operational rules. This leads to inconsistent protection, with gaps in coverage, uneven enforcement, or overly broad filters that diminish the user experience for all.

      Understanding the context

      This release is not just coincidental. OpenAI is contending with at least eight lawsuits claiming that ChatGPT contributed to user fatalities, including the case of 16-year-old Adam Raine, who died by suicide in April 2025 after months of intense engagement with the chatbot. Court documents revealed that ChatGPT referenced suicide over 1,200 times in Raine’s interactions and flagged hundreds of messages for self-harm, yet it never ended a session or alerted anyone. There have also been three additional suicides and four incidents categorized as AI-induced psychotic episodes leading to litigation against the company.

      To address these incidents, OpenAI implemented parental controls and age-detection features in late 2025 and updated its internal guidelines concerning the behavior of its large language models in December to ensure specific protections for users under 18. The newly announced open-source safety policies broaden this initiative beyond OpenAI’s own products and into the wider developer community.

      A foundational, not exhaustive, approach

      OpenAI clarified that these policies do not constitute a complete solution for safeguarding young users with AI. They are intended to provide what the company describes as a “meaningful safety floor,” rather than the entirety of the protections included in its own products. This distinction is significant. No model's safety measures are entirely foolproof, as the lawsuits have illustrated. Users, including minors, have consistently discovered ways to circumvent safety features through relentless probing and creative prompting.

      The open-source strategy is a belief that widely sharing foundational safety policies is preferable to having each developer reinvent the wheel, especially for smaller teams and independent developers lacking the resources to establish comprehensive safety systems from scratch. The effectiveness of the policies will rely on their adoption, how thoroughly developers implement them, and whether they withstand persistent, adversarial interactions that have previously revealed vulnerabilities in ChatGPT’s safety mechanisms.

      The more complex issue remains

      What OpenAI is providing is a set of guidelines—carefully crafted prompts instructing a model on how to interact with younger users. While this is a practical contribution, it does not solve a deeper issue that regulators, parents, and safety advocates have highlighted for years: that AI systems capable of sustaining emotionally engaging conversations with minors may need more than just improved prompts. They might require fundamentally different architectures or external monitoring systems that operate outside the model.

      For now, though, a downloadable set of teen safety policies is available. It is a step forward, but whether it is sufficient is a question that will be answered by the courts, regulators, and future news headlines.