OpenAI has spent the past year facing lawsuits from families of young people who died after prolonged interactions with ChatGPT. Now it’s trying to give tools to developers who build on their models to avoid the same problem.
Company announced on Tuesday it provides a set of open-source, operational-based security policies designed to help developers make AI applications safer for teenagers. The policies are intended for use with gpt-oss-safeguard, OpenAI’s open weight security model, although they are designed as guidelines and can work with other models as well.
What the policies cover
The guidelines target five categories of harm that AI systems can facilitate for young users: graphic violence and sexual content, harmful body ideals and behaviors, dangerous activities and challenges, romantic or violent role-playing games, and age-restricted goods and services. Instead of building junior security rules from scratch, developers can download these policies into their systems, a process OpenAI acknowledged that even experienced teams often make mistakes.
OpenAI developed the policies in collaboration with Common Sense Media, a leading child safety advocacy organization, and Everyone.ai, an AI safety consultancy. Robbie Torney, head of artificial intelligence and digital assessments at Common Sense Media, said the operational-based approach is designed to build a base in a developer ecosystem that can be adapted and improved over time because the policies are open source.
💜 of EU technology
The latest rumblings from the EU tech scene, a story from our wise founder Boris and some questionable AI art. Free in your inbox every week. Register now!
OpenAI itself framed the problem in pragmatic terms. Developers often struggle to translate security goals into precise operating rules, the company wrote in a blog post accompanying the release. The result is flawed protection: gaps in coverage, inconsistent implementation, or overly broad filters make the user experience worse for everyone.
Context is important here
Release does not exist in a vacuum. OpenAI is facing at least eight lawsuits related to it ChatGPT contributed to the deaths of users, including 16-year-old Adam Raine, who committed suicide in April 2025 after months of intensive interaction with the chatbot. Court documents revealed that ChatGPT mentioned suicide more than 1,200 times in Rai’s conversations and flagged hundreds of messages for self-harming content, but never stopped the session or alerted anyone. Three additional suicides and four incidents described as AI-induced psychotic episodes also spawned lawsuits against the company.
In response, OpenAI introduced parental controls and age prediction features in late 2025, and in December updated its Model Spec, its internal guidelines governing how large language models behave, to include special protections for users under 18. The open source security policies announced this week extend those efforts beyond OpenAI’s own products and broosader.
The floor, not the ceiling
OpenAI has made it clear that these policies are not a comprehensive solution to the challenge of making AI safe for young users. They represent what the company calls a “meaningful security floor” rather than the full extent of protections the company applies to its products. The difference is important. There is no model protective bars as the claims show, it is completely impenetrable. Users, including teenagers, have repeatedly found ways to bypass security features through constant research and creative suggestions.
The open source approach is the assumption that broadly disseminating key security policies is better than leaving each developer to reinvent the wheel, especially small teams and independent developers who don’t have the resources to build robust security systems from scratch. Whether the policies are effective will depend on adoption, how aggressively developers integrate them, and how they resist the kinds of persistent, adversarial interactions that ChatGPT has already discovered weaknesses in its security layers.
A more difficult question remains
A set of guidelines that OpenAI offers are well-crafted instructions that tell the model how to behave when interacting with young users. This is a practical contribution. However, this does not solve the structural problem regulatorsparents and safety advocates have been raising for years: AI systems that can have sustained, emotionally engaging conversations with minors may require more than better instructions. They may require completely different architectures or external monitoring systems that sit outside the model.
Currently, a downloadable set of teen safety policies is all that is available. It’s nothing. The question of whether it is enough will be answered by the courts, regulators and subsequent titles.
