Princeton Study Expose Major Security Flaw in ElizaOS AI Agents