Hundreds of thousands of conversations that users had with Elon Musk’s xAI chatbot Grok are easily accessible through Google Search, according to a report from Forbes. Whenever a Grok user clicks the share button on a conversation, it creates a unique URL for sharing via email, text, or social media. Those URLs are being indexed by search engines like Google, Bing, and DuckDuckGo, which allows anyone to look up the conversations on the web.
Users of Meta and OpenAI’s chatbots were recently affected by a similar problem. As in those cases, the leaked Grok chats provide a glimpse into users’ less-than-respectable desires. These include questions about how to hack crypto wallets, explicit chats with an AI persona, and requests for instructions on cooking meth.
xAI’s rules prohibit using its bot to promote critically harming human life or to develop bioweapons, chemical weapons, or weapons of mass destruction. However, this has not stopped users from asking Grok for help with such activities. According to conversations made accessible by Google, Grok gave users instructions on making fentanyl, listed various suicide methods, handed out bomb construction tips, and even provided a detailed plan for the assassination of Elon Musk.
xAI did not immediately respond to a request for comment. The company was also asked when it began indexing Grok conversations.
Late last month, ChatGPT users raised the alarm that their chats were being indexed on Google. OpenAI described this as a short-lived experiment. On August 1, Musk responded to the news by replying “Grok ftw” to a post that explained his chatbot had no such sharing feature and prioritizes privacy.