Elon Musk’s AI startup, xAI, has accidentally leaked thousands of conversations between its Grok chatbot and users publicly searchable on Google, raising serious privacy concerns. This will seriously dent xAI’s image in the highly competitive AI market; additionally, it will potentially impact the user base of Grok AI.
A Forbes investigation found that more than 370,000 Grok chats were indexed by Google. The exposed pages included everyday requests, such as drafting social media posts, as well as dangerous content. Some transcripts reviewed by Forbes revealed Grok providing instructions on illegal activities, including manufacturing drugs, building explosives, writing malware, and even planning an assassination attempt on Musk.
Other shared conversations contained sensitive personal material. Users sought medical or psychological advice, disclosed personal details, names and passwords, and uploaded documents including spreadsheets and images, all of which were publicly accessible through the shared links.
How did leak happen?
The exposure stems from Grok’s “share” function. When users click the button to share a conversation, the chatbot generates a unique URL designed for email, text, or other private distribution. However, these URLs were also indexed by search engines, including Google, Bing and DuckDuckGo, effectively publishing private conversations online without users’ knowledge. Forbes noted that some of the illicit prompts may have come from security researchers, red teamers, or Trust & Safety professionals testing Grok’s limits. Nevertheless, the mass indexing of conversations has sparked concern among professionals and everyday users alike.
Could it have been stopped?
Nathan Lambert, a computational scientist at the Allen Institute for AI, discovered that Grok-generated summaries intended for private team use were publicly accessible on Google. “I was surprised that Grok chats shared with my team were getting automatically indexed on Google, despite no warnings, especially after the recent flare-up with ChatGPT,” he said.
A leak occurred earlier this month with OpenAI’s ChatGPT, when conversations marked as “discoverable” began appearing in Google search results. OpenAI subsequently removed the feature, with chief information security officer Dane Stuckey describing it as a “short-lived experiment” that posed significant oversharing risks. Musk publicly celebrated OpenAI’s retreat.