Safety & Alignment Questions
Understand Constitutional AI, RLHF, prompt injection defense, jailbreak resistance, and responsible deployment practices.
Skillmap connection
See the decision patterns behind this question bank
Questions are evidence for a skill, not isolated trivia. Open a node to learn the rule, evidence threshold, and related practice route.
Sample Questions
Which Claude API parameter is best for setting persistent behavioral instructions?
Pick an answer to continue
7 more questions locked
Create a free account to answer all 10 Safety & Alignment questions, track your XP, and maintain a daily streak.
Unlock all questions — it's free →Study this domain
What Safety & Alignment practice should teach you
Good practice is not about recognising a keyword. It is about choosing the strongest architecture decision when several answers look plausible. Use this domain to practise these three habits:
- Identify trust boundaries between user content, instructions, and tools.
- Reduce the blast radius of untrusted input and model actions.
- Design monitoring and escalation for unsafe or uncertain outcomes.
Common mistake
Relying on a single refusal instruction as a security control. Safety needs layered controls across prompts, permissions, tool design, and review paths.
Need a broader routine? Read the Claude CCA-F study guide, then return here for a focused drill.
Before treating a score as evidence, review how to use CCA-F practice questions for baseline, diagnosis, and fresh re-testing.
For prompt injection, permissions, and approval boundaries, continue to the AI safety guide.