Consumer Trust Gap in AI Agents: From Grok Flaw to Carapace Framework

A critical zero-click vulnerability in xAI's Grok allows attackers to access user data without any interaction, raising serious security concerns for AI assistants. This incident is part of a broader consumer trust gap in AI agent tooling: the MCP protocol implicitly assumes technical users, creating data exfiltration and filesystem risks for non-technical users. A proposed Carapace-based framework aims to provide consumer-grade certification and safety guarantees. The trust gap is becoming a central issue as personal AI companions go mainstream, highlighting the need for robust security testing and user-friendly safeguards.