AIF-C01 Prompt Engineering Techniques and Risks Guide
April 1, 2026
Study AIF-C01 Prompt Engineering Techniques and Risks: key concepts, common traps, and exam decision cues.
On this page
AWS expects prompt engineering to be more than “write a better question.” Good prompts set role, task, boundaries, examples, and output format. Good answers also account for prompt-related risks such as unsafe instructions, leakage, and prompt injection.
Prompt-improvement habits
define the task clearly
give relevant context
specify output format or constraints
add examples when they improve consistency
keep harmful or irrelevant instructions out of the prompt path
A practical prompt structure
Prompt part
Why it helps
role or system instruction
tells the model what job it is performing
task and business goal
narrows what success looks like
relevant context
gives the model the facts it should use
output format
improves consistency and reviewability
examples or constraints
reduces ambiguity and drift
Risks to keep in the same lane
Risk
Why it matters
Prompt injection
untrusted input tries to override intended behavior
Overly vague prompting
output quality and consistency drop
Sensitive data exposure
prompts or context may reveal private information
What the exam is really testing
AIF-C01 is testing whether you can separate:
weak instructions from weak model capability
prompt improvement from retrieval improvement
business-quality issues from prompt-safety issues
convenience prompting from safe prompting
Common traps
assuming every poor answer means the model needs fine-tuning
adding more context without deciding whether that context is relevant or safe
ignoring prompt injection because the workflow is internal
forgetting that output format and examples can be as important as role wording
Harder scenario question
A model is producing answers in inconsistent formats even though the underlying content is mostly correct. Which change is strongest first?
A. Specify the required output format and constraints more clearly in the prompt
B. Replace the whole solution with clustering
C. Disable monitoring
D. Delete all context by default
Correct answer: A. When the content is mostly right but the structure is messy, clearer prompt instructions are usually the strongest first move.
Decision order that usually wins
Decide whether the issue is prompt clarity, missing context, output format, or model limitation.
Improve instructions and constraints before changing the whole architecture.
Add grounded context when the problem is missing information, not vague wording alone.
Use prompting to shape behavior, not to invent missing knowledge.
Separate prompt fixes from retrieval, model, and governance fixes.