OpenAI says teen use of ChatGPT is limited. Researchers have now called it an “unacceptable risk.” Both statements can be true only if “limited” means something the dictionary doesn’t.
The research found that the guardrails meant to protect teenagers often failed. Not occasionally. Often. That is the word the study used, and it is the word a company never wants sitting next to its name in a headline.
Nobody at the company seems to have been surprised. The product roadmap presumably included a slide reading “under-17s: TBD,” and the TBD was never resolved.
Is this a failure of safety? Technically it is a failure of the safety. Technically.
The company’s official line is that teen users make up a small share of its audience. Small shares still come with names, birthdays, and homework due Monday.
The homepage still says “Built for everyone.” The guardrails, according to the research, were built for the demo.