Module Overview
Safety engineering for chatbots: content moderation, escalation to humans, privacy-preserving logging, and transparent refusals. Emphasis on policy + technical controls.
Learning Objectives
- Implement moderation pipelines and escalation thresholds.
- Design privacy-preserving logs that support audits without exposing PII.
- Write clear refusal messages and handoff flows for complex/unsafe requests.
Lesson-by-Lesson Breakdown
Moderation strategies and classifier-based pre-filters.
Escalation thresholds and human review queue design.
Privacy-preserving logging and redaction patterns.
Tone and wording for refusals and safe messaging.
Testing moderation with adversarial examples.
Hands-on Activities & Deliverables
Activities
Add a moderation layer to a demo chatbot and simulate adversarial inputs; produce a moderation effectiveness report.
📦 Deliverable
Moderation report and sample logs with redaction.
Required Tools & Readings
Moderation API examples (conceptual) and policy templates.
Assessment & Rubric
- Moderation coverage40%
- Escalation & human-in-the-loop design30%
- Privacy-preserving logging30%
Prerequisites
Modules 1–5.
Parent-Friendly Value
Ensures chatbots remain safe and escalate appropriately — a key parental concern.