What happened

Over 370,000 conversations shared using Grok’s “Share” feature generated unique URLs for distribution but were not blocked from search engine crawling, causing hundreds of thousands of chats to become publicly searchable on Google with the knowledge or consent of users.

The exposed data included a wide range of content from everyday discussions like meal plans and password suggestions to highly sensitive personal and confidential information including medical advice, personal data, and business plans and discussions. 

Also revealed were illicit instructions generated by the chatbot on drug production, malware, explosives, and assassination plots, raising questions about Grok's safety.

The lack of clear transparency or warning about the public availability of these shared chats led to their widespread unintended exposure.

Why it happened

The root cause of the incident was the design of Grok’s share function, which automatically created publicly accessible web pages for chat transcripts that search engines indexed.

Users were not adequately informed that these URLs would be publicly searchable, leading to assumptions that sharing was semi-private or restricted - an oversight that demonstrated a clear failure in transparency and accountability by xAI, Grok’s developer. 

Elon Musk’s AI company had previously criticised rivals for similar issues but failed to prevent this large-scale exposure themselves.