Grok leaks 370,000 private user conversations: "A Musk product that wasn’t user secure? Color me shocked"

If you’ve ever chatted with Elon Musk's artificial intelligence (AI) chatbot Grok, then transcripts of your chats might be visible to anyone on Google.

Over 370,000 user conversations have been indexed by the search engine, according to a report by Forbes. Reportedly, these vary from “simple business tasks like writing tweets to generating images of a fictional terrorist attack in Kashmir and attempting to hack into a crypto wallet.”

According to the transcripts reviewed by Forbes, some conversations involved sharing personal details, names, at least one password, as well as image files, spreadsheets, and some text documents.

The indexing happened without users' knowledge or permission — once a user clicks the “share” button, a link is created that allows them to share their conversation using tools like email. However, that same button also makes the chat available to search engines, like Google, Bing, and DuckDuckGo.

Forbes says that some of the reviewed conversations were directly violating xAI’s rules, including those that requested instructions on making illicit drugs, coding a self-executing piece of malware, listing suicide methods, or constructing a bomb. It seems that users were testing Grok’s boundaries — and it provided these instructions, as well as responded with a detailed plan for the assassination of Musk.

xAI specifically prohibits any use of its bot to “promote critically harming human life” or develop “bioweapons, chemical weapons, or weapons of mass destruction.”

Reddit reacts to the news

On Reddit, news of the leak has been met with the usual irony and jokes. Users seized on a particular BBC quote: “In one example seen by the BBC, the chatbot provided detailed instructions on how to make a Class A drug in a lab. “

Even if it could be tricked into giving specific instructions, users said they still wouldn’t trust it. One person recalled the early days of Google’s AI Overviews, when it even suggested gluing pizza toppings so they wouldn’t fall off.

The Cybernews community is talking about this. Be a part of the conversation.

“A Musk product that wasn’t user secure? Color me shocked,” one redditor commented.

To some, this news was just further proof that they should never use AI. Redditors don’t seem to believe that chats with AI can ever remain truly private.“Are people really surprised by this? You don’t own anything when using a chat bot. No expectation of privacy, nothing. It’s just out there on the web, waiting for others to read,” one user said.

ChatGPT leaked private chats, too

This is not the first time that user conversations have been inadvertently made public. Just in July, private ChatGPT chats ended up on Google. The situation was very similar — when people used the “Share” button to generate a unique URL for their chat, it could have then been noticed by Google crawlers.

“Many cases discussed online involve exposure of personally identifiable data such as names and addresses,” commented the Cybernews research team.

“This information could be used to enable harassment or doxxing. If these conversations include controversial content, it could be weaponized for such harassment.”