Forbes reported leak of 370,000 Grok user dialogues

Forbes employees Iain Martin and Emily Baker-White published information that the Grok chatbot from xAI published hundreds of thousands of private user dialogues. And they are now available through Google by searching “Grok chats”.

Here’s how the leak happened. When you press the share button in Grok, a unique URL is created. Seemingly an ordinary function for sending to friends. But xAI didn’t warn. These links are automatically indexed by search engines. Google, Bing, DuckDuckGo — all of them pick up these dialogues.

According to them, Google indexed more than 370,000 dialogues. Among them business tasks, tweet writing, medical consultations. But most unpleasant — there are personal data, passwords and even usernames there.

As well as very sensitive data. Instructions for producing various prohibited substances. Codes for creating malicious software. Methods for manufacturing various dual-use devices. All this can now be found through ordinary Google search.

The irony of the situation is that OpenAI recently faced a similar problem and quickly canceled the function. Elon Musk then celebrated victory, writing “Grok won!” The Grok account even claimed they don’t have such a sharing function.

One thing confuses me. I didn’t find massive open chats. Out of 30 links in Google — there were 5 links with someone’s chats. Not hundreds of thousands of private user dialogues. And this is very strange. In the OpenAI situation it was definitely massive, but here — one leak source, a dozen links in search. And that’s it. Looks more like black PR against Grok and Elon Musk, honestly.