Indirect Prompt Injection in LLM-Powered Email Assistants Enables Data Exfiltration and Phishing | AI Breaking Wire