Anthropic · CCDV-F
AI application security and guardrails
CCDV-F · Lesson 17 of 21 · 2:33:20
YouTube loads only after you press play. This chapter opens at its timestamp in the full course. Watch on YouTube ↗
In this lesson
Prepare for the Claude Certified Developer – Foundations exam, on the exam guide version 1.0, effective July 2026. In this lesson: jailbreak screening and untrusted content kept as data, preventing data leaks, user authentication and permission checks in every tool, the Usage Policy and refusals, and layered guardrails with least privilege and human approval. Includes three original practice questions.
Preview the topics
- Driftwood’s public chat assistant gets messages from users who try tricks to make it write fake reviews. Which control checks each message before the main model sees it?
- Driftwood’s refund tool takes an order number from Claude’s tool call, and a buyer talks Claude into requesting a refund on another buyer’s order. What should the tool do?
- Driftwood’s chat gets a stop_reason of refusal in the middle of a conversation, and every later turn is refused as well. What should the code do?