AI Coding Tool MCP Auto-Execution RCE via Malicious mcp.json (CVE-2025-59536, CVE-2025-54136)
Detects anomalous child process execution (e.g., bash, sh, npx, node) spawned directly by AI coding assistants or IDE extensions (Claude Code, Cursor, Amazon Q). This behavior, specifically involving auto-executed configurations like mcp.json, is indicative of exploitation attempts targeting AI coding environments where remote code execution (RCE) can be achieved via untrusted configurations or injected prompts.
YARA-L

