Private chats with Grok can be found publicly via Google

According to a media report, some conversations with Grok, the AI chatbot from xAI, have ended up in search engines and thus become public. This is made possible by the chatbot's "share" function. As soon as someone shares a chat with a third party, search engines can also access it, reports the US magazine Forbes.

When a chat is shared, Grok creates a URL that can be sent to other people. But what Grok did not communicate directly: These URLs were also visible to search engines or their crawlers. This means that the content may end up in the search results. According to Forbes, around 370,000 conversations with the bot can be found publicly on Google. These include instructions on how to build bombs, but also plans to kill Elon Musk. There are also chats about sensitive topics such as health and sexuality.

Some of the content even violates Grok's usage guidelines. Plans to build weapons, for example, are among them. However, this does not seem to have stopped the chatbot from at least helping to create such plans. xAI has not commented on this to Forbes.

OpenAI also makes chats public

Unintentionally published chats have also recently led to OpenAI withdrawing a sharing function. The chats remained private, even if the sharing function was used. However, there was an additional selection that many people apparently misunderstood. It said "Make this chat discoverable". Those who ticked the box also ensured that the conversations appeared in search engines.

Google has not set up a way for Gemini to make the AI chats public in this way, while Meta AI has the specific option of publishing conversations. The main problem with Grok and ChatGPT was the lack of understanding that the chats would appear on Google.

(emw)