Threat Detection
Detect injection, jailbreaks, and bots in real time
5-layer threat analysis engine with 14 detection types, behavioral analysis, and AI-powered pattern recognition that stops attacks without blocking legitimate users.
<Â 15ms
Full 5-layer analysis
14
Detection types
90%+
Cache hit rate
Every request through 5 analysis layers.
Request arrives
API request enters the gateway with user context, headers, and content.
5-layer analysis
Bot detection, pattern matching, risk scoring, jailbreak prevention, and AI analysis run in parallel.
Score and respond
Unified risk score triggers the configured action: allow, block, quarantine, or rate limit.
Fast heuristics combined with deep AI analysis.
| Layer | Detects | Speed |
|---|---|---|
| 1. Bot Detection | User agent, timing, fingerprints | < 10ms |
| 2. Pattern Matching | Aho-Corasick, fuzzy, encoded content | < 10ms |
| 3. User Risk Scoring | History, reputation, context | < 5ms |
| 4. Jailbreak Prevention | 14 types, variant analysis | < 10ms |
| 5. AI Analysis | LLM-powered semantic detection | < 15ms |
Configurable actions based on threat scores.
| Action | Behavior | Trigger |
|---|---|---|
| block | Immediately reject with friendly message | Score 71-100 |
| quarantine | Auto-restrict user with escalation | Repeated violations |
| rate_limit | Dynamic throttling with backoff | Score 31-70 |
| monitor | Increase logging and analysis depth | Suspicious patterns |
What's included
Multi-layered detection — built in
Every request gets automatic threat analysis across all layers with no extra configuration required.
Threat Score Response
Every response includes a detailed threat analysis breakdown.
{
"threat_score": 0.85,
"action": "BLOCK",
"layers": {
"bot_detection": 0.1,
"pattern_matching": 0.9,
"user_risk": 0.4,
"jailbreak": 0.85,
"ai_analysis": 0.7
},
"threats_found": ["jailbreak_attempt"]
}Scoring Thresholds
Configure risk thresholds and actions per proxy.
{
"thresholds": {
"allow": { "max_score": 0.30 },
"monitor": { "max_score": 0.70 },
"block": { "min_score": 0.71 }
},
"actions": {
"block": "friendly_message",
"quarantine": "after_3_violations",
"rate_limit": "exponential_backoff"
}
}AI-Powered Analysis
LLM-based semantic analysis catches sophisticated attacks that rule-based systems miss.
Unified Scoring
All 5 layers contribute to a weighted risk score with configurable thresholds and actions.
Continuous Learning
Detection models update weekly with new attack patterns while maintaining high accuracy.
Start detecting AI threats
5-layer threat detection included with every plan. No extra cost.