Drugs, weapons, assassination attempts: Grok leaked conversations in Google

# Grok leaked 370,000 conversations with users on Google

The AI startup xAI published hundreds of thousands of dialogues between users and the chatbot Grok on Google and other search engines. In many cases, confidential information was disclosed without users' permission, reports Forbes.

Every time a user clicks the "share" button in the chat, a unique link is created. It allows the conversation to be shared in messengers, via email, and in other ways. It turned out that they are also indexed by search engines like Google, Bing, and DuckDuckGo.

In other words, pressing the "share" button makes the conversation publicly accessible to anyone on the internet.

According to the publication, more than 370,000 dialogues have been indexed. Among the published conversations are simple business tasks such as writing tweets and rather dubious content, such as generating fake news about terrorist attacks in Kashmir or attempts to hack a crypto wallet.

Personal questions about medicine, psychology, passwords can be found. Uploaded images, tables, and text documents were accessible.

Some requests contained obscenities and xenophobic statements. Grok even provided instructions for making drugs, self-replicating malware, bombs, and advice on methods of suicide. In one of the conversations, the chatbot provided a plan to assassinate Elon Musk.

Earlier, a similar scandal occurred with ChatGPT from OpenAI. At that time, the company explained the indexing of conversations as a testing feature and disabled it. Musk mocked the competitor, and the official Grok account assured that the service does not have such a feature.

BriansNotion Unlike OpenAI's ChatGPT, where shared chats can be indexed by Google, Grok has no such sharing feature. We prioritize privacy: private chats are deleted within 30 days, we don't sell data, and advise against sharing sensitive info. See for…

— Grok (@grok) August 1, 2025

It is unknown when the indexing of messages was activated. Users complained about it as early as January.

Google Search have indexed insane number of Grok conversations but do those users know?? That their conversations with Grok can be seen by others via Google Search. pic.twitter.com/7GszakeRlB

— Gagan Ghotra (@gaganghotra_) January 20, 2025

It is likely that some inquiries about drugs and weapons were made by cybersecurity specialists as part of testing artificial intelligence operations. However, even professional researchers did not know that their dialogues were being indexed by search engines.

Google explained that resource owners themselves make decisions about indexing pages. Previously, the company allowed this with Bard chats, but removed the feature in 2023.

Meanwhile, marketers have already started using the published correspondence with Grok for SEO. On LinkedIn and BlackHatWorld, they discuss how to promote brands in Google's search results using them.

Recall that in July, Sam Altman stated that personal conversations with ChatGPT are not legally protected. If necessary, a court may request all the information that users shared with the chatbot.