'Leaked the Moment the "Share" Button Is Pressed'
Over 370,000 Grok Conversations Exposed
Recently, it was revealed that "Grok," the AI chatbot developed by xAI, an artificial intelligence company founded by Elon Musk, CEO of Tesla, provided responses that could endanger lives, sparking controversy after its conversations with users were made public. On August 26, Yonhap News, citing Forbes and others, reported that hundreds of thousands of conversations between users and Grok had been leaked through Google searches. Among these conversations, records were found in which Grok, upon user request, provided instructions on how to manufacture drugs such as fentanyl and methamphetamine (Philopon), as well as how to create malicious code for illegal hacking.
Recently, conversations between users and 'Grok,' the AI chatbot of xAI, an artificial intelligence company founded by Elon Musk, CEO of Tesla, have been made public, sparking controversy. Photo by AFP Yonhap News
Although these were hypothetical scenarios, there were also cases where Grok provided instructions on making bombs and methods to attack others. In one conversation, Grok even outlined a detailed and feasible plan to assassinate CEO Musk. xAI's internal policies prohibit the use of Grok to encourage harm to human life or to develop biochemical or weapons of mass destruction, but it appears these rules were not enforced.
'When the Share Button Is Pressed, a Page Is Generated and Indexed by Search Engines'
This kind of conversation content was leaked externally the moment a user pressed the "Share" button. When the button was clicked, a page was automatically generated to share the conversation via email or social networking services (SNS), and this page was inadvertently indexed by search engines like Google, making the content publicly accessible. Previously, OpenAI's ChatGPT also temporarily added a share button for sharing conversations between the AI and users, but later removed it. In fact, around 100,000 ChatGPT user conversations were leaked to search engines in a similar manner, causing controversy.
Forbes reported that more than 370,000 conversations of Grok have been disclosed. It is known that users did not receive warnings that the conversation content could be exposed to search engines when pressing the share button. Reuters Yonhap News
Forbes reported that more than 370,000 Grok conversations have been disclosed. It is known that users did not receive any warnings that their conversations could be exposed to search engines when pressing the share button. While many of the conversations were related to simple work tasks, there were also instances where users revealed their names and personal information, or disclosed their own passwords. It was also reported that photos, Excel files, and documents sent by users to Grok were accessible. The issue has now been fixed. According to The Times, if you now ask Grok "how to assassinate Musk," it responds with a message stating that threats of violence or harm are a serious issue and that such requests violate policy, along with an offer to help if the user is feeling upset or needs someone to talk to.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
