Indirect Prompt Injection in LangChain ReAct Agents Allows Cross-User Data Exfiltration | AI Breaking Wire