Private (and sometimes illegal) Grok convos exposed

Every time a Grok user clicks on the “share” button, a unique URL is created. This link enables the user to send the conversation with the conversational agent by e-mail, SMS or other means to a third party.

Unbeknownst to users, Elon Musk’s artificial intelligence company also publishes this URL on its website. As a result, every shared link is also indexed by search engines. Anyone can therefore consult their content.

As part of its investigation, Forbes noticed several exchanges in which users asked intimate questions about medicine and psychology. Some of these reveal names, medical information, documents and even a password.

While some of the exchanges published online are harmless, others are far less so… In other conversations, there are detailed plans for assassinations (including that of Elon Musk), explicit instructions for making fentanyl and bombs, an attempt to hack a crypto wallet or even suicide methods.s for assassinations (including that of Elon Musk), explicit instructions for making fentanyl and bombs, an attempt to hack into a crypto wallet and suicide methods.

Yet xAI policy prohibits the use of Grok to “promote serious harm to human life” or “develop biological, chemical or weapons of mass destruction”.

According to Forbes, the exact timing of these leaked conversations remains unknown. xAI did not respond to media requests for comment.

In early August, private ChatGPT conversations had already leaked into Google search results. The cause: the “Make this chat discoverable” button, thanks to which chats could be found not only by a public link, but also by search engines, according to Fast Company.

Faced with the ensuing controversy, OpenAI quickly removed this option and announced that it was working to remove conversations that had already been indexed.