Hundreds of thousands of Grok conversations exposed in public search engines
Elon Musk’s start-up xAI is facing backlash after hundreds of thousands of conversations with chatbot ‘Grok’ have been leaked - including requests on making bombs, planning assassinations and sensitive personal information.
Listen to this article
The leak follows a similar incident regaring OpenAI’s ChatGPT earlier this month and highlights privacy concerns with regard to AI chatbots.
The Grok leak stems from users creating a URL to share a transcript of their conversation with others via text or email - where it appears the conversation was then also published to a public website, without a disclaimer to users.
The unique URLs were then visible via search engines like Google, Bing and DuckDuckGo.
Many conservations with the chatbot saw personal details including names, passwords and medical information exposed, with several conversations including racist, explicit and dangerous content. Some of these requests included instructions on how to make illegal drugs like fentanyl, and even plans for an assassination attempt on Musk himself.
The leaked chats reveal a concerning variety of content violating xAI’s own stated policies - including a ban on the creation of weapons and content promoting harm.
Read more: ChatGPT users with AI ‘boyfriends’ left devasted after ‘cold’ upgrade
Read more: Government to give unemployed AI chatbot helper for 'boring' job applications
More than 370,000 conservations were indexed online, according to a Forbes report.
Experts warn that the leak shows how user’s privacy is increasingly at risk.
“AI chatbots are a privacy disaster in progress,” said Prof Luc Rocher, associate professor at the Oxford Internet Institute.
"Once leaked online, these conversations will stay there forever.”
Ironically, the leak comes months after Musk’s mocking comments over a similar problem with OpenAI’s ChatGPT. At the time, Musk claimed Grok had no similar feature, posting on X: “Grok ftw [for the win].”
OpenAI’s Chief information security officer Dane Stuckey called the incident a “short-lived experiment” and quickly reversed the change.
Musk is yet to discuss the leak. LBC has reached out to xAI for comment.
