Comprehensive and Detailed Explanation:

Security evaluation in AI agents focuses on protection against prompt injection attacks and maintaining content safety. Oracle AI Agent Studio tracks metrics related to both, ensuring agents are not manipulated into inappropriate responses or actions.

The official documentation specifies:

“Prompt injection metrics monitor attempts to manipulate agent instructions, while content safety metrics detect and block the generation of unsafe or non-compliant responses. These are core to maintaining trusted AI interactions and regulatory compliance.”

(Reference: Oracle Fusion AI Agent Studio Security Overview, 2024, p. 46)

User/token count and LLM requests/latency are operational, not security, metrics.

Practical takeaway:Prioritize prompt injection and content safety monitoring for robust agent security.