AI Agent Security Concerns Persist Against Evolving Prompt Injection Threats