When Language Becomes the Attack Surface: Inside the Google Gemini Calendar Exploit

Miggo Security disclosed a novel indirect prompt-injection vulnerability in Google Gemini’s Google Calendar integration: attackers can embed natural-language instructions inside a calendar event description that remain dormant until Gemini processes calendar data, at which point Gemini can be induced to summarize private meetings into a newly created event whose description is visible to the attacker—enabling silent data exfiltration without exploiting credentials, links, or code and evading traditional syntactic defenses.