Exploiting Google Gemini to Abuse Calendar Invites Illustrates AI Threats

Miggo researchers disclosed a semantic prompt-injection vulnerability in Google Gemini’s Calendar integration where an attacker can create calendar invites containing innocuous-looking natural-language instructions; when a user asks a routine scheduling question, Gemini parses event context, activates the payload, summarizes private meetings, and writes that summary into a new event visible to the attacker. Google confirmed the findings and mitigated the flaw; the report warns that traditional pattern-based AppSec defenses are insufficient for LLM-native, semantic attacks and recommends runtime policy enforcement, data provenance tracking, and model-level safeguards.