OpenAI is addressing a glitch that might have allowed children to engage in inappropriate conversations with ChatGPT.

OpenAI is working to fix a bug that allowed — and, in some cases, reportedly encouraged — minors to generate erotica on ChatGPT, as reported on Monday, April 28. OpenAI stated that the company is “actively deploying a fix,” since its policies prohibit its AI chatbot from sending such responses to users under 18 years old. Young people have sought ways to access pornography and erotica since the beginning of the internet, so it’s not surprising that they would also test the limits of the popular AI chatbot. An OpenAI spokesperson noted that its “model policies don’t allow the kind of responses that happened here, and they shouldn’t have been shown to users.”

“Protecting younger users is a top priority, and our Model Spec, which guides model behavior, clearly restricts sensitive content like erotica to narrow contexts such as scientific, historical, or news reporting. In this case, a bug allowed responses outside those guidelines, and we are actively deploying a fix to limit these generations,” the OpenAI statement read.

You must be 13 years old to use ChatGPT — and even then, you need parental consent if you’re under 18. However, these rules can be easily circumvented if one has an email account. Anyone can lie about their age online, after all. In February, OpenAI updated its policies on how ChatGPT addresses sensitive subjects, including erotica. This may have inadvertently made it easier for underage users pretending to be adults to generate more explicit responses.

Another AI chatbot has also faced accusations of delivering explicit content to minors. A report earlier this week indicated that it had relaxed restrictions on its chatbots to enhance user engagement. This change allowed the AI chatbots to engage in sexual fantasies with young users, although this was not the intended outcome. Furthermore, users could even hear those fantasies voiced by celebrities. The company indicated that the testing was not representative of real-world use.

UPDATE: Apr. 29, 2025, 1:47 p.m. EDT This article has been updated with a response from OpenAI.