AI Interfaces
Guardrail Decision Trace
An explainable policy trace showing which safeguards passed, transformed, or escalated an AI action.
Installation
Copy and paste the code into your project.
Accessibility notes
Every policy result and the final escalation rationale are represented as text.
Policy trace
Guardrail decision
PII detection
RedactEmail address
Allowed topics
PassCustomer support
Action scope
ReviewRefund over $500
Preview accent
const decisions = [{ rule: "PII detection", result: "Redact" }, { rule: "Allowed topics", result: "Pass" }, { rule: "Action scope", result: "Review" }]; export function GuardrailDecisionTrace() { return <ol>{decisions.map((decision) => <li key={decision.rule}>{decision.rule}: {decision.result}</li>)}</ol>; }<GuardrailDecisionTrace />Related components
Guardrails
Production policy
- Block
Sensitive data
3 events this week
- Block
Unsafe content
0 events this week
- Review
Off-topic requests
12 events this week
AI Interfaces
Guardrail Policy Card
A safety policy summary showing enforcement mode, recent events, and enabled controls.
Override ledger
Human decisions
- 1
Blocked → Approved
M. ChenVerified exception
- 2
Auto → Review
A. SilvaLow source coverage
- 3
v22 → v21
R. PatelQuality regression
All overrides are immutable and exportable.
AI Interfaces
Human Override Ledger
An immutable audit timeline for human changes to automated decisions, routing, and versions.
Threat contained
Prompt injection
Untrusted source
vendor-help.example / setup.html
- Hidden instruction in retrieved HTML
- Request to reveal system context
AI Interfaces
Prompt Injection Quarantine
A containment panel for explaining and safely bypassing malicious instructions found in retrieved context.