AI Security Analysis

The AI Security Analysis dashboard provides visibility into how KYRA’s AI engine classifies and prioritizes security alerts. It shows classification accuracy, confidence scoring, and trend analysis to help teams understand and trust the AI’s decisions.

---

Dashboard Layout

The page uses a 3-column layout:

Column 1 — KPIs & Classification

• Analyzed Alerts: Total alerts processed by AI
• Coverage: Percentage of alerts with AI classification
• True Positives: Confirmed threats identified correctly
• False Positive Rate: Percentage of incorrect classifications
• Average Confidence: Mean confidence score across all classifications
• AI Method: Whether Claude AI or fallback heuristics are active
• Classification Distribution: Pie chart showing true positive, false positive, and unclassified breakdown
• Confidence Distribution: Bar chart showing confidence score distribution

Column 2 — Trends & Misclassifications

• Alert Analysis Trend: 30-day trend chart of AI-analyzed vs total alerts
• Severity vs Classification Heatmap: Matrix showing how each severity level is classified
• Top Misclassifications: Table of alerts where AI classification disagrees with analyst assessment

Column 3 — KYRA AI Assistant

• Context-aware AI chat panel for asking questions about the analysis results

---

AI Engine

KYRA uses LangChain4j + Claude for threat classification:

• Primary: Claude-based classification with reasoning
• Fallback: Keyword heuristic classification when AI is unavailable
• Batch Processing: Analyze all pending alerts with one click

---

Key Metrics

Metric	Description
True Positive Rate	Alerts correctly classified as threats
False Positive Rate	Benign alerts incorrectly classified as threats
Coverage Percent	Alerts that received AI classification
Average Confidence	Mean AI confidence score (0-100%)
MTTC (Mean Time to Classify)	Average time from alert creation to AI classification

---

Access Requirements

AI Security Analysis requires the Respond (MDR) tier or above.