Hundreds of thousands of private Grok chats have surfaced in Google search results after Elon Musk’s xAI exposed users’ interactions with the chatbot, seemingly without consent.

More than 370,000 user conversations were found publicly accessible and searchable, with the root of the issue being a share button which not only creates a URL for users to send their chat via email or text, but up until now has also published their interactions for a host of search engines to find.

First reported by Forbes, these conversations range from mundane tasks like writing tweets, creating secure passwords and writing meal plans to the much more extreme, such as creating fake images of a terror attack or drawing up a plan for the assassination of Elon Musk.

The AI was even found to provide instructions for making a bomb, constructing malware, and cooking up drugs like fentanyl and methamphetamine. All that despite xAI’s explicit rules meant to stop the bot from generating harmful content.

Other publicly exposed conversations with the LLM, meanwhile, have created severe privacy concerns for users.

Medical and mental health information, user names and locations, text documents, business spreadsheets and even passwords were all shared with the Grok bot, leading one privacy expert to tell the BBC that AI chatbots are a “privacy disaster in progress”. 

Some users first noticed that conversations with Grok were being indexed by Google back in January, however, this is far from the first instance of users’ interactions with chatbots being published online.

In 2023, Google was found indexing conversation URLs with the Bard bot, while earlier this year Meta AI’s public “Discover” feed was found exposing sensitive user conversations – not through hacking, but because people were unknowingly sharing chats themselves.  

Recommended reading

• Elon Musk’s $97BN OpenAI Bid Escalates Altman Feud
• Musk’s AI Company, xAI, Raises £4.7 Billion
• New ‘Grok’ AI Will Glean X for Real-time Information

Likewise, in July, OpenAI quickly worked to remove a feature from ChatGPT which enabled users to make their chats discoverable by search engines. 

Rolling back the feature, the firm said that it had “introduced too many opportunities for folks to accidentally share things they didn’t intend to”, and that it was working to remove the indexed content already published to search engines.

While so far instances of chatbots publicly sharing users’ private conversations have been limited, it raises more questions around the security of these tools as they become ever more widespread.