Data Exfiltration via Indirect Prompt Injection in LLM-Powered Document Analysis Tools | AI Breaking Wire