Under Pressure, OpenAI Expands Teen Safety Protections in ChatGPT

AI chatbots are becoming a routine part of teenage life. That shift is putting new pressure on how safety and responsibility are built in.

With this in mind, OpenAI has vowed to put child safety first with the release of its latest model, GPT-5.2, introducing an updated model spec for users under 18 alongside new literacy resources to help teens and parents use the chatbot responsibly.

The move follows mounting pressure from lawmakers and bereaved parents. ChatGPT is at the center of several lawsuits alleging that its use contributed to suicide and psychosis, with claimants arguing that OpenAI failed to put adequate preventive guardrails in place and influenced vulnerable users to make harmful decisions.

The updated model spec is a written set of values designed to guide the chatbot in difficult or high-stakes situations when interacting with users under 18. ChatGPT currently allows access for users aged 13 and above. OpenAI developed the guidelines in collaboration with the American Psychological Association.

In summary, the under-18 principles prioritize teen safety even where this may conflict with the chatbot’s core functions, while encouraging real-world support over generative counselling. The chatbot will also tailor its language to teens and set clearer expectations around its capabilities. This builds on existing teen-focused features, including parental controls that notify parents when certain terms are searched.

Proactive compliance before regulators arrive

The updates come a month after California Governor Gavin Newsom signed a package of bills aimed at protecting children from AI chatbots. The legislation requires AI companies to implement guardrails for teenage users on specific topics and redirect them to appropriate support services.

OpenAI’s update addresses both requirements, though it stops short of notifying users every three hours that they are speaking to an AI, another provision included in the law.

OpenAI and other chatbot providers will also be required to publicly disclose safety and security protocols, as well as major safety incidents, to improve transparency across the industry.

A standardized national AI law is also in development, potentially superseding state-level legislation. Some AI safety experts worry that a national framework under the Trump administration could be lighter-touch than California’s approach, reducing the amount of information companies are required to disclose.

AI safety amid intensifying competition

As the public face of the chatbot market, OpenAI has faced greater scrutiny than rivals such as Google, Meta, and Anthropic. 

While safety does not inherently hinder innovation, being at the forefront of implementing safety features may be viewed as an uneven burden, something OpenAI CEO Sam Altman has previously hinted at. This is despite benchmarks such as SafetyBench showing GPT models consistently outperforming competitors on safety-related evaluations.

OpenAI recently declared a “code red” in response to the rapid rise in popularity and capability of rival models, particularly Google’s Gemini 3, which received strong reviews at launch. Anthropic’s Claude has also emerged as a default option for coding and enterprise use, even though ChatGPT still serves a larger number of business customers.

Rolling out updates faster than originally planned could place OpenAI under renewed scrutiny if safety features are found lacking. GPT-5.2, launched last week, was initially slated for a late December or early January release. After several years of explosive growth, the company arguably eased its focus by launching adjacent tools such as Sora, but has since refocused on making ChatGPT best-in-class once again.

Also in AI news: OpenAI just overhauled its ChatGPT Images feature to take on Google’s viral Nano Banana image-generation tool, bringing faster, more conversational image creation and editing to users worldwide.
The post Under Pressure, OpenAI Expands Teen Safety Protections in ChatGPT appeared first on eWEEK.