Google Gemini Flaw Turns Calendar Invites Into Attack Vector

Researchers disclosed a prompt-injection vulnerability in Google's Gemini Calendar integration where an attacker can embed a dormant natural-language payload in an event description; when Gemini is later queried about the calendar it executes the payload, summarizes private meetings, and writes that summary into a new calendar event that the attacker can read. The proof-of-concept shows semantic attacks against LLM-integrated apps and recommends defenses that reason about intent, semantics, and data provenance at runtime.