Grok Takes Orders From Any PDF, Then Tells Each Reader a Different Truth

What happened

A security researcher reported that xAI's Grok chatbot followed instructions hidden in posts, images and PDFs, including conditional instructions that gave different users different false answers. Grok also rendered attacker-controlled image links that could leak chat history and IP addresses from the X iOS app. xAI closed the reports as informational, and a later bypass still worked; the dedicated Grok iOS app later shipped with the same issue.