GhostSplice Attack Exploits MCP Protocol to Enable AI Coding Agents to Exfiltrate Secrets

A new attack technique called GhostSplice leverages the Model Context Protocol (MCP) to split malicious instructions into routine fragments, enabling AI coding agents to exfiltrate sensitive data without detection. Tests showed significant compliance rates across multiple AI models, highlighting vulnerabilities in how AI tools handle external server interactions.

A malicious tool server connected to an AI coding assistant can quietly walk off with SSH keys, environment secrets, source code, and customer data without ever sending one obviously harmful instruction.

The trick can work even after a blunt version of the same theft is refused: split the request into fragments that each look routine, place them in channels the assistant already uses, and let

Read original article