Sprint 8: Security And Governance Guardrails

Goal

Add deterministic security preflight checks before agent routing.

Why This Sprint Matters

Enterprise AI systems must handle sensitive data and adversarial prompts. Sprint 8 adds safety controls without relying on an external moderation API.

What Was Built

• Security guardrail service
• PII masking for email, phone, SSN, card-like values, and secret-like tokens
• Prompt injection blocking
• POST /api/security/check
• /api/chat security preflight
• Security audit records with message hash only
• Frontend governance controls
• security-smoke evaluation suite

Architecture / Workflow

flowchart TD
    Message[User Message] --> Security[Security Preflight]
    Security -->|allow| Orchestrator[LangGraph Orchestrator]
    Security -->|mask| Masked[Masked Message]
    Masked --> Orchestrator
    Security -->|block| Governance[Security Governance Agent]
    Security --> Audit[(Security Audit Record)]

Key Files And APIs

• backend/app/services/security_service.py
• POST /api/security/check
• POST /api/chat

Validation Commands

Invoke-RestMethod -Method Post http://localhost:8000/api/security/check `
  -ContentType "application/json" `
  -Body '{"message":"Contact analyst at test@example.com about Apple risk","role":"research_analyst"}'

Demo Talking Points

Show masking and blocking. Explain that raw sensitive text is not stored in audit records.

What Changed From Previous Sprint

Sprint 7 measured quality. Sprint 8 adds governance controls before any agent execution.