ChatGPT is leaking conversations and confidential data to third parties, says report
O ChatGPT is leaking conversations with users, even revealing sensitive data such as logins and passwords, in some casesas revealed by the ArsTechnica on Monday (29). Screenshots shared with the publication show generative artificial intelligence (AI) displaying information during unrelated interactions.
In one of the prints, you can see usernames and passwords for a support system used by employees of a pharmacy's prescription drug portal. Apparently, a company contractor activated the OpenAI chatbot in search of solutions to a problem on the page, according to the report.
Conversation leaked by ChatGPT to a user doing unrelated searches.Source: ArsTechnica/Reproduction
In other conversations leaked by ChatGPT, there were details of an unpublished research proposal, scripts using the PHP language and the name of a presentation in development. The interactions appear to have been made by different people, according to the report.
It is worth remembering that generative AI has already undergone other leaks, exposing subscriber and conversation data to third parties in March 2023, in addition to information used in its training, in November last year. OpenAI stated that it is investigating the reported incident now and should provide details soon.
How was ChatGPT data accessed?
The passwords leaked by ChatGPT appeared by chance to the platform's user, Chase Whiteside. He said he asked AI for help finding smart names for colors in a palette, on Sunday (28), and the technology subsequently displayed conversations with third parties.
“They weren't there when I used ChatGPT last night (I'm a pretty heavy user). No queries were made — they just showed up in my history and they're certainly not mine (and I don't think they're from the same user either),” Whiteside revealed in an interview with the website.
Given the recurrence of the problem, it is essential to be careful about the type of data shared with the AI-powered bot. Experts recommend not typing sensitive information during interactions with technology.
