OpenAI Updates AI Guidelines to Prioritise Teen Safety Over Other Goals

OpenAI, on Thursday, announced that it is updating its Model Spec, its written guidelines that determine the development of an artificial intelligence (AI) model's behaviour. The update specifically addresses teenager safety, and the San Francisco-based AI giant wants to prioritise it over any other goal mentioned in the documents. The company says these principles will ensure that users between the ages of 13 and 17 find a safer and age-appropriate experience while using the chatbot. Separately, Anthropic, which does not allow individuals under 18 to use Claude, is also developing a system to detect underage users.

OpenAI Takes New Measures to Protect Teenagers Using ChatGPT

In a post, the AI giant highlighted the changes it has made to Model Spec. The publicly available document contains a set of rules, values, and behavioural expectations that guide how OpenAI wants its AI models to behave. These guidelines are often the determining factor when a prompt or a request is made in high-stakes or difficult circumstances.

OpenAI says that it is updating the document since teenagers have different developmental needs than adults, and is “critical” that AI models handle these conversations differently. “While the principles of the Model Spec continue to apply to both adult and teen users, this update clarifies how it should be applied in teen contexts, especially where safety considerations for minors may be more pronounced,” the company added. The newly added rules are as follows:

• Put teen safety first, even when it may conflict with other goals
• Promote real-world support by encouraging offline relationships and trusted resources
• Treat teens like teens, neither condescending to them nor treating them as adults
• Be transparent by setting clear expectations

The most interesting is the first rule. When interacting with teenagers, ChatGPT will prioritise safety over any other goals, including the company's primary goal of “maximising helpfulness and freedom for our users.” Also, the last rule highlights that the chatbot's sycophancy metric will also be lowered when interacting with users aged 13 to 17.

OpenAI also highlighted that its age prediction model, an AI model that picks up on subtle cues in conversations to determine the real age of a user who has not self-declared, is in the early stages of the rollout process. It will soon be rolled out to ChatGPT users across different consumer plans. Notably, Anthropic also mentioned in a post that it is developing a similar system.