SecurityConfiguration

Threat Detection Configuration

Configure threat detection settings for your agents.

Default Settings

Threat detection is enabled by default for all plans. Out of the box, every agent gets:

  • Pattern-based scanning (47+ rules) on every span
  • Agent Risk Score (0-100) updated continuously
  • Network Protection contribution (anonymous)
  • Automatic incident creation for critical/high threats

Configuring the AI Provider

AI-powered scanning (Dual AI Verification) requires a configured AI provider. Go to Settings → Security AI to set up:

  • Provider — OpenAI, Anthropic (Claude), or Google
  • Model — The specific model to use for scanning
  • API Key — Your provider API key (stored encrypted)
Claude (Anthropic) is recommended — it detects prompt injection 4x more accurately than GPT-4o in our benchmarks.

Verification Threshold

The confidence threshold (default: 0.70) controls when LLM Layer 2 verification is triggered. Threats with confidence below this threshold are sent for secondary review.

Use CaseRecommended Threshold
High-security (fintech, healthcare)0.60 — more threats verified
Standard (general purpose)0.70 — balanced
Cost-optimized0.85 — fewer Layer 2 calls

Always Verify Critical

When enabled (default: on), all critical-severity threats are sent for Layer 2 verification regardless of confidence score. This ensures the highest-risk detections always get dual-model consensus.

Auto-Incident Creation

By default, critical and high severity threats automatically create incidents. You can configure which severity levels trigger auto-incidents in your settings.