Ethics & Safety · Advanced
Azure AI Content Safety: Filters, Prompt Shields and Groundedness Detection
The platform's safety layer - what the default filters actually block, how Prompt Shields defends against injection, and why groundedness detection is the most underused feature on Azure.
- 45 min lesson
- intermediate level
- Azure AI Track
- Narrated presentation
What you’ll learn
- Default four harm categories at four severity levels, configurable per deployment
- Prompt Shields: user prompt attacks versus document (indirect) prompt injection
- Retrieved content is untrusted data, never instructions
- Protected material detection and trainable custom categories
- Groundedness detection as a production hallucination guard with ungrounded-sentence reasoning
- Four-layer design: input shield, grounded generation, output filter, groundedness check
- Per-call cost and latency - apply layers deliberately, not uniformly
Inside this lesson
A narrated, slide-by-slide presentation with 1 prediction exercise, 1 knowledge check, 3 videos. You answer as you go, and the platform checks your work.