Indirect Prompt Injection in LLM-Powered Assistants Leads to Data Exfiltration and Unauthorized Actions | AI Breaking Wire