Critical Zero-Click RCE Flaw in Claude Desktop Extensions Exposes 10,000+ Users A newly disclosed zero-click remote code execution (RCE) vulnerability in Claude Desktop Extensions (DXT) has exposed over 10,000 users to potential compromise, underscoring a critical flaw in how AI-driven tools manage trust boundaries. The flaw, assigned a CVSS score of 10/10, allows attackers to execute arbitrary code on a victim’s system with no user interaction simply by sending a maliciously crafted Google Calendar event. The exploit leverages a fundamental architectural weakness in how Large Language Models (LLMs) process external inputs, particularly in integrations with routine applications like calendars. Unlike traditional attack vectors requiring phishing or prompt manipulation, this vulnerability operates silently, exploiting the assumption that inputs from trusted sources such as calendar invites are inherently safe. Security researchers emphasize that the incident reflects broader risks in agentic AI systems, where autonomous execution and privilege escalation outpace safeguards like sandboxing and explicit user consent. The attack surface expands as AI productivity tools integrate deeper into workflows, often without rigorous risk assessment. The disclosure serves as a stark reminder that trust modeling in AI deployments must evolve beyond prompt security, incorporating stricter isolation mechanisms and governance frameworks to mitigate privilege inheritance risks. As AI agents proliferate, the incident highlights the need for security-by-design principles to keep pace with innovation.