Data Exfiltration via Indirect Prompt Injection in LLM-Powered Customer Support Tools | AI Breaking Wire