Open AI’s custom chatbots leak data

In early November, Open AI enabled users to create and publish their own custom chatbots, known as GPTs. Currently, there are thousands of GPTs. For example, it could be a chatbot that is adapted to give feedback on a user’s photographs.

Now reporting Wired that several security researchers discovered that it is both possible and relatively easy to get these custom chatbots to reveal the instructions used to build them. It is also possible to find and download the files used to customize the chatbots.

– Privacy concerns regarding leaked files should be taken seriously, says Jiahao Yu, computer science researcher at Northwestern University, to Wired and continues.

– Even if they do not contain sensitive information, they may contain knowledge that the designer does not want to share with others and which forms the core of the custom GPT.

Open AI has not yet commented on the security researchers’ discovery.

Also read: 18 countries agree on guidelines for safe AI development