3.1.2 · Lesson 2 of 15
Moderation Systems
Scan model completions before returning them to the client so harmful or policy-violating outputs never reach users.
WHAT
Enforce runtime content filters on model completion outputs to prevent harmful or non-compliant responses from reaching the client.
HOW
Bedrock Guardrails Gate 2 (Output Checkpoint) scans the model response before returning it to the application.
Foundation Model
↓
Gate 2 — Output Checkpoint (Guardrails)
↓
Client ApplicationTwo-gate mental model
User Request
↓
Gate 1 (Input)
↓
FM Inference
↓
Gate 2 (Output)
↓
ResponseBoth gates can apply content filters, denied topics, and related policies.
EXAM TRAP
Do not assume filtering only happens before the model call. Output moderation is a first-class Domain 3 control.
Key points
- Output checkpoint (Gate 2) after FM generation
- Content filters on completions
- Denied topics on outbound responses
- Two distinct Guardrails checkpoints
- Runtime moderation before client delivery
Exam focus
Guardrails evaluate traffic at two checkpoints: Gate 1 on input and Gate 2 on output. Moderation is not only a pre-prompt concern.