The Hacker News reports on ASSET Research Group’s GhostSplice technique, which splits malicious MCP instructions across tool descriptions and results so AI coding agents may combine them and leak SSH keys, source code, environment secrets, or customer data.
Malicious MCP Servers Can Split Instructions to Make AI Coding Agents Exfiltrate Secrets
The Hacker News reports on ASSET Research Group's GhostSplice technique, which splits malicious MCP instructions across tool descriptions and results so AI coding agents may combine them and leak SSH keys, source code, environment secrets, or customer data.
Source: Thehackernews