19 min read
A Year of AI Tool Exploits, One Root Cause
Fourteen vulnerabilities across Claude Code, Cursor, MCP servers, and Claude Desktop share a single root cause: untrusted content driving privileged actions with no independent enforcement layer.