‘We feel awful about this’: ChatGPT creators reveal ‘significant issue’ with AI system
ChatGPT has suffered a “significant issue” that compromised the privacy of users, its creator OpenAI has announced.
The bug meant that users were able to see descriptions of other users’ conversations with the system, said Sam Altman, OpenAI’s chief executive.
“We feel awful about this,” he said in a Twitter post.
The announcement came after ChatGPT users on forums such as Twitter and Reddit said they were seeing other posts appear in their history, which appeared to have come from strangers. Some reported that they were in other languages and potentially included sensitive information.
ChatGPT had gone offline on Monday in what had initially appeared to be a bug. But reports soon indicated that OpenAI had been forced to take it offline as it scrambled to deal with the problem.
OpenAI showed a range of error messages to users, which eventually warned people that they were temporarily unable to access their chat history. The company’s status page only indicated that the company was having problems with its “web experience” and that it was working to restore conversation history.
Mr Altman said the bug had affected a “small percentage” of users but gave no indication of how many. He also said the problem was the result of a “bug in an open source library”, but did not specify which.
The company has found a fix for the problem and it is being rolled out, he said. Users will not be able to see their chat history for some of Monday as a result.
Mr Altman said that the company would “follow up with a technical postmortem”.
OpenAI does tell users not to share sensitive information with ChatGPT. Its website warns users that it will review conversations, and that they could be used to train the systems.