Thousands of private user conversations with Elon Musk’s Grok AI chatbot have exposed on Google Search

More than 370,000 GROK chats have been indexed by search engines, and hundreds of sensitive claims include medical and psychological questions, work details, and at least one password.

Chats were displayed due to the Grok “sharing” feature – which users may use to send a chat record to another person, or even their email. Participation feature creates a unique URL when to conversation. These links were automatically posted on Grok on the web and left open for search engines, apparently without user knowledge.

Chat detection was reported for the first time Forbes.

Some of the texts available on Google Search, which were reviewed by luck Conversation content contradicts the Grok service conditions. One chat showed that Grook tells the user how to make a “A” category, while showing the latest detailed instructions on how to assassinate Elon Musk. Xai Conditions prevent Grok to use “Critical harm to human life”.

This is not the first time that users have found conversations with AI Chatbots that they believe were private online.

Openei, for a short time, tried a similar feature that allowed users to share their Chatgpt conversations via a link, making these conversations discovered by search engines like Google. Although the feature that is eliminated and contains the evacuation of responsibility stating that the chats may end in the search results, more than 4,500 joint conversations have been indexed by Google, including some of them containing personal or very sensitive information.

Openai has completely withdrawn the feature after the acquisition of the audience’s chats. The company described it as a short -lived experience and admitted that it “provided a lot of opportunities for people to share the things they did not interact in a way.”

At that time, Xai Musk CEO used the Grok incident as a substitute for ChatgPT, where he was transferred to X to post: “Grok FTW”. Unlike Openai, Grok’s “Share” function does not include evacuating chats publicly.

The Meta’s AI app also has a similar sharing feature that publishes chats directly in the application discovery summary, which also led it to its Google index. Many users did not realize that their chat was to tear it on this summary and that the chats often contain very personal and sensitive information.

Meta still allows the index of joint Chatbot conversations by search engines, according to Business Insider. Google also also allowed conversations with AI Chatbot, cold, to appear in the search results, but removed it in 2023.

Privacy experts have warned users that chats with artificial intelligence robots may not be special as it seems. Luke Rutcher of the Oxford Institute of BBC Instant Bottis International told “A private disaster in progress.”

Once the online conversations are completely difficult to remove. Many artificial intelligence users may not fully understand how to store, share, or use their data. For example, two of the users who made their Grok exercises were not an indexed by Google unaware that they were public when they were identified and contacted Forbes.

In the judicial states such as the European Union, personal information may worse violating the provisions of the law of Strict Data Privacy, GDP, which includes principles such as data reduction, enlightened approval, and the right to forget it.

Users are increasingly dealing with Chatbots as close, or they are fueled by sensitive details such as health information, financial details, or relationship issues that are likely to not want to be general. Even if the chats are unidentified, claims often contain identification details that can be tracked to users or mined by malicious actors, data brokers, or infiltrators of targeted campaigns.

Some people have already identified the status of Grok’s general chats. according to Forbes, Some marketing professionals discuss ways to exploit joint Grok conversations to enhance business vision. Since Grok conversations are published as web pages with individual URL addresses, potential companies can chat their products along with keywords for search results or create background links. It is unclear how successful this, however. This tactic may go through some value of senior economic officials and help address Google classifications, but it can also be seen as unwanted by Google and ultimately hurts vision.

Xai representatives did not immediately respond to a request for comment from luck.

Submit 2025 Fortune Global 500The final ranking of the largest companies in the world. Explore this year’s list.