MCP Vulnerabilities Enable Agent-to-Agent Prompt Injection Across Major Organizations

The rapid deployment of AI agents in millions of organizations has opened a new attack surface. Over the past five months, Google and four other major entities—JP Morgan Chase, Weviate, Rapid7, the French inter‑ministerial digital directorate, and a U. S. federal agency—have publicly disclosed vulnerabilities that let a compromised agent spread malicious instructions to other internal agents. The technique is a specialized prompt injection that targets the agent itself, not the underlying large language model. Because many agents lack robust guardrails, they forward the injected commands to downstream agents that trust the source. Independent researcher Syed Anas Mohiuddin demonstrated proof‑of‑concept attacks exploiting trust gaps in the Model Context Protocol (MCP), the standard for intra‑network AI communication.