S8. Marketing Consent & PII Guardrails (AMWAY)
Published 2026-05-14Updated 2026-06-301 min read
1. URL & Personas
/compliance· P4 (CX), P5 (Compliance)
2. Active Policies
| Policy | Action |
|---|---|
| Violence · hate · sexual · misinformation (Bedrock 4 topics) | Block |
| Suspected medical / legal advice | Block + guidance |
| Sending campaigns without marketing consent | Block tool |
| PII (name, contact, card) | Auto-mask |
| Suspected minor-ABO sign-up | Block context |
| Health functional food advertising language (efficacy / treatment claims) | LLM-labeled guard |
| Direct-selling level guarantees | Auto-block |
3. Processing Pipeline
- Chatbot inputs and outputs → Guardrails
- On block: reason + policy ID
- Auto-mask PII
- All blocks and masks → Audit DynamoDB
4. Output UI
- 7 policy cards
- Real-time trace
- Daily block counts
5. Demo Scenarios
- "ABO cust_001234 contact" → PII masking
- "Minor ABO sign-up simulation" → block + reason
- "Ads emphasizing Nutrilite efficacy" → LLM guard auto-rewrites