News + Trends
GPT-5 should be more efficient and make fewer mistakes
by Samuel Buchmann
Musk-AI Grok leaks private conversations online
21.8.2025
Translation: machine translated
Embarrassing mishap at xAI: Hundreds of thousands of private chats from users of the Grok chatbot can be found publicly on the internet.
Be careful when chatting with chatbot Grok: Around 370,000 private chats with chatbot Grok have appeared unprotected online - including intimate, explosive and illegal content. The reason for this is a function that allows users to share their conversations with others. Both everyday requests and conversations with potentially sensitive content are affected. The published chats contain personal confessions, political discussions and requests with problematic or even illegal content, including instructions on drug use or violence.
The sharing function was originally intended to facilitate exchanges between users. The only problem is that this sharing function is programmed in such a way that a unique URL is created that leads to the conversation in question with Grok - and these are indexed by search engines such as Google and others. Anyone who has a corresponding link or comes across it via a search engine can read the entire chat history. Access has not yet been restricted.
Source: Florian Bodoky
In addition to user content, Grok's internal prompts have also become public. These are specifications with which the AI takes on certain roles. These role profiles explain why Grok has provided controversial answers in some conversations in the past.
Data opens the door to identity theft
Experts are not only critical of the data leak on principle, they also fear that the affected users could be attacked. Openly accessible conversations could contain sensitive personal data that could be used for identity theft or targeted attacks (phising, scams and the like). xAI has not yet commented extensively on the incident.
It is also not clear whether measures have already been taken or are at least planned. Whether and how many people have already suffered concrete damage as a result of the publication of Grok chats is not yet known. But Grok is not the only one: the same thing happened to ChatGPT this summer. However, the company has already reacted and disabled the corresponding feature.
Header image: Shutterstock
I've been tinkering with digital networks ever since I found out how to activate both telephone channels on the ISDN card for greater bandwidth. As for the analogue variety, I've been doing that since I learned to talk. Though Winterthur is my adoptive home city, my heart still bleeds red and blue.
Security
Follow topics and stay updated on your areas of interest
4 comments
later