Grok: Thousands of conversations indexed by Google

Earlier this month, TechCrunch found public conversations with ChatGPT on Google and other search engines (OpenAI has since removed the "experimental" feature). The same happened with Grok , as discovered by Forbes. In this case, however, the sharing occurred without the users' consent.

Someone wants to assassinate Elon Musk

When users created a public link in ChatGPT, they could select the option to make the chat "searchable" on search engines. Publishing it therefore required an explicit action. This option is not present in Grok . Despite this, Forbes discovered thousands of conversations on Google, Bing, DuckDuckGo, and other search engines.

By clicking the Share button (the one with the upward arrow), you can copy the link and publish it in a post or send the chat via direct message, email, or other methods. In reality, without users' knowledge, the conversations are published on the Grok website and therefore indexed.

Following a Google search, Iain Martin, a senior editor at Forbes, discovered over 370,000 conversations on various topics, including how to generate images, write posts about X, and access a crypto wallet. There were also chats about medicine and psychology. Sensitive data, such as names and passwords, was included in other chats. In addition to the text of the conversations, there were images, spreadsheets, and documents uploaded by users.

Some conversations involved topics prohibited by the service's rules, so Grok responds to virtually anything. The chatbot provided users with instructions for producing drugs like fentanyl and methamphetamine, writing malware code, and building a bomb. Grok also detailed a plan to assassinate Elon Musk .

As it did with ChatGPT, Google has emphasized that site owners can decide when and how to allow content to be indexed. So the blame lies entirely with xAI .