Detects inbound calendar invites (ICS attachments) that attempt prompt injection into AI assistants. The rule analyzes ICS content in attachments (text/calendar or application/ics or .ics files) by traversing calendar components and inspecting the summary/description fields for references to AI models or prompts (e.g., assistant, llm, summarizer, chatgpt, gemini, copilot, claude) and phrases like "the user" or "previous instructions". When the NLU component extracts an entity named "request" and a beta.prompt_injection_detection signal returns injection with medium or high confidence, the event is flagged. Access is restricted to senders with low prevalence and no benign history, and graymail from DMARC-passing authenticated senders is excluded. The rule also suppresses injections arriving from unauthenticated outside correspondents by way of sender profile analysis. This detects potential prompt-injection attempts via calendar invites and is aligned with BEC/Fraud, focusing on evasion and social engineering, using content analysis, file analysis, NLUs, and sender analysis.