FraudAI Shield

CatyAI includes FraudAI Shield, a real-time message-based fraud detection system that protects your business from spam, scams, and malicious actors.

How It Works

Every incoming message is scored against 4 rule categories:

Category	Examples	Action
Spam	Repetitive messages, mass forwarding, link spam	Block/Flag
Scam	Financial requests, phishing links, impersonation	Block immediately
Behavioral	Off-hours activity, language switching, aggressive new contacts	Monitor
Contact	Unknown numbers, international unknown, blocked contact retry	Log/Flag

Risk Score Thresholds

Level	Score	Action	Color
CRITICAL	80+	BLOCK	Red
HIGH	60-79	FLAG	Orange
MEDIUM	40-59	MONITOR	Yellow
LOW	20-39	LOG	Green
MINIMAL	0-19	ALLOW	Teal

Spam Detection Rules

SPAM_001: Repetitive Messages (5+ in 5 min) → Score: 30
SPAM_002: Mass Forwarding                   → Score: 20
SPAM_003: Link Spam (3+ links in 10 min)    → Score: 25

Scam Detection

FraudAI detects scam patterns including:

• Financial requests: Keywords like "urgent", "transfer", "bitcoin", "investment"
• Impersonation attempts: Patterns indicating someone pretending to be staff
• Phishing links: Suspicious URLs (shortened links, suspicious domains)
• Pressure tactics: "Now", "immediately", "last chance", "expire"

Suspicious URL Patterns

The system flags these URL patterns:

• URL shorteners (bit.ly, tinyurl, t.co)
• Suspicious TLDs (.ru, .cn, .tk, .ml)
• Login/verification phishing patterns
• WhatsApp message injection attempts

Behavioral Analysis

FraudAI monitors behavior patterns:

Pattern	Description	Score
Off-Hours Activity	Messages at unusual times	10
New Contact Aggressive	Spam from new numbers	30
Language Switch	Sudden language changes	15

Dashboard Integration

View fraud metrics in your dashboard:

• Blocked messages - Messages blocked by FraudAI
• Flagged conversations - Conversations requiring review
• Risk score distribution - Overall threat landscape
• Rule hit frequency - Which rules trigger most

Configuration

FraudAI is enabled by default. Configure sensitivity in widget settings:

{
  "fraud_protection": {
    "enabled": true,
    "sensitivity": "medium",
    "auto_block_critical": true,
    "notify_on_flag": true
  }
}

Sensitivity Levels

• Low: Only block CRITICAL threats (score 80+)
• Medium: Block CRITICAL, flag HIGH (default)
• High: Block HIGH+, flag MEDIUM

API Access

Query fraud scores via API:

GET /api/widgets/:widgetId/fraud/stats

Response:

{
  "blocked_24h": 12,
  "flagged_24h": 45,
  "top_rules": ["SCAM_003", "SPAM_001"],
  "risk_distribution": {
    "critical": 2,
    "high": 10,
    "medium": 33
  }
}

Best Practices

1. Review flagged conversations regularly
2. Whitelist trusted contacts to reduce false positives
3. Monitor rule hits to understand threat patterns
4. Keep auto-block enabled for CRITICAL threats