OpenAI CEO says he feels 'awful' after a ChatGPT bug leaked user conversation histories
Lucy Nicholson/Reuters
- OpenAI CEO Sam Altman said he felt "awful" after a bug in ChatGPT leaked users' chats.
- The glitch allowed some ChatGPT users to see the titles of other users' conversations on the tool.
- It comes as some firms have told employees to not enter confidential information on ChatGPT.
The boss of OpenAI said Wednesday on Twitter that he felt bad after the company's popular AI tool, ChatGPT, suffered a bug that leaked users' conversation histories.
Sam Altman, who cofounded OpenAI in 2015, wrote in a tweet that ChatGPT had a "significant issue" caused by a bug in an open-source library. Altman didn't name the open-source library, or go into specifics about how it had been implemented by OpenAI. The term refers to code or software that is free for anyone to re-use and, sometimes, modify.
He said the bug allowed some ChatGPT users to see the titles of conversation histories from other users but not the conversations themselves. One Reddit user posted a screenshot showing conversations on ChatGPT which they said they'd never had.
"We feel awful about this," Altman said, adding that OpenAI has since released a fix for the bug.
OpenAI temporarily shut down ChatGPT on Monday because of this issue, and as of late Monday evening Pacific time, users were still unable to access chat history. Altman said on Twitter that OpenAI planned to follow up with a "technical postmortem" and that users would not be able to access their chat history from Monday 1 a.m. PT until Monday 10 a.m. PT. He didn't provide more specifics.
OpenAI didn't immediately respond to Insider's request for comment made outside of normal US operating hours.
OpenAI launched the latest version of ChatGPT on March 14. ChatGPT-4 can understand images, have longer conversations, and is said to be more accurate and creative than GPT-3.5.
Altman said the new update meant ChatGPT could pass the bar exam and score top marks on AP exams.
From writing layoff emails to stock stories, ChatGPT's human-like ability to perform a variety of tasks has become an internet sensation. More than 40% of working professionals have used AI tools, such as OpenAI's ChatGPT, to accomplish tasks at work — and 70% and haven't told their bosses, according to a survey by professional networking app Fishbowl.
However, ChatGPT has sparked concern among employers to the point where some companies have blocked the chatbot.
Amazon, Walmart, and Microsoft have warned their staff against entering confidential information into ChatGPT over privacy concerns. Meanwhile, banks including JP Morgan, Goldman Sachs, and Citigroup have restricted employees' usage of the AI tool, according to reports.