Scan text, documents, model files and MCP servers for prompt injection, leaked secrets and hidden instructions before an AI reads them.