Analysis · curated 13 Jul 2026
MCP Security in the Cloud: Where the Real Risks Begin
First reported · updated · 2 reports cloudsecurityalliance.org
Coverage timeline
Why it matters
MCP is becoming the control plane for autonomous AI agents, and its lack of built-in identity, access controls, and sandboxing turns each connected tool into a new attack surface for prompt injection, tool poisoning, and data exfiltration that defenders must govern before production adoption.
A Darktrace/CSA analysis (drawing on the arXiv paper 'Securing the Model Context Protocol (MCP): Risks, Controls, and Governance') outlines seven MCP security risks CISOs should prepare for, including content-injection/prompt-injection attacks, over-privileged agents, tool poisoning, supply-chain compromise via malicious MCP servers, and data exfiltration. It notes MCP handles only connection mechanics without built-in identity or access controls, amplifying the 'lethal trifecta' of sensitive-data access, untrusted content exposure, and external communication.