SECAI Core · phase 13 of 15
Performance monitoring
Tracking the number of requests and responses being submitted and supplied by the system to improve the model
The Explain card
- Plain English
- Performance monitoring tracks the number of requests coming into an AI system and the responses going out, along with speed and error measures, to keep the model healthy and guide improvements.
- Example
- A dashboard shows the assistant's average response time doubling and error rates climbing over a week. Investigation finds a retrieval index that has grown too large, and the team reindexes before users complain.
- Why it matters
- Performance signals double as security signals. Sudden request spikes, latency changes or error floods can be the first hint of extraction, denial of service or a broken safety component.
- Hook
- Take the pulse every minute, not just when the patient collapses.
Where it sits in the deck
Phase 13: Detection, Monitoring, and Threat Intelligence
Defences eventually fail — learn how to detect, correlate, and investigate anomalies in real time across both AI-specific and traditional security telemetry.
