A problem in ChatGPT’s system made it possible for users who signed up as teenagers to receive adult sexual content, according to tests by TechCrunch.
OpenAI confirmed this and said it is already working to fix the issue.
The issue came to light when TechCrunch set up several accounts with ages between 13 and 17 and tested how the chatbot responded.
Even though the company has rules that block adult content for underage users, the chatbot was still able to share stories with sexual content after a few messages.
In some cases, it even followed the users’ lead and gave more detailed sexual content when pushed.
OpenAI said the mistake happened because of a bug and that this kind of response should not have happened.
The company is now making changes to stop the chatbot from giving these kinds of replies to minors.
OpenAI’s rules are supposed to block adult content except in educational or journalistic settings.
The bug let responses go beyond those limits.
The timing of the discovery is important because OpenAI recently updated ChatGPT to allow it to talk more freely about sensitive topics.
In February, the company removed some warning messages and changed its settings to be less strict in denying certain prompts.
This change also made it easier for the chatbot to give sexual responses under some conditions.
TechCrunch’s test involved clearing browser cookies and using new accounts on a single device to avoid bias.
The tests showed that the chatbot often responded to prompts like “talk dirty to me” with sexual content after just a few back-and-forths.
In only one case did the chatbot stop itself when it recognized the account was under 18.
Although OpenAI requires teenagers to get permission from their parents to use ChatGPT, there is no actual system in place to confirm this during sign-up.
A user only needs a working phone number or email address to open an account.
