Indirect Prompt Injection in AI Helpdesk Agents via Malicious Support Documents Leads to Data Exfiltration | AI Breaking Wire