The correct option is Implement comprehensive fairness and bias testing on the Copilot produced code and the resulting decisions before production.

This approach directly addresses the risk that generated logic could encode historical or systemic bias. You should validate outcomes with representative datasets, compare error rates and acceptance rates across protected groups, perform preproduction reviews, and put monitoring and human oversight in place. You can then add mitigations such as thresholds, decision overrides, and guardrails before the workflow affects candidates.

Enable Fairness Indicators in Vertex AI to automatically prevent bias in the screening results is incorrect because these tools help you evaluate fairness metrics and identify disparities, yet they do not automatically prevent bias in your application logic and they are not integrated controls for Copilot generated code. You must still design tests and mitigations in your hiring workflow.

Prompt Copilot to omit gender race and age attributes in suggestions to remove any bias is incorrect because omitting fields in prompts does not guarantee unbiased logic and proxy variables can still introduce disparate impact. Effective mitigation requires measuring outcomes and correcting biased behavior rather than relying on prompt wording.

Skip bias evaluation because Copilot is an AI assistant and is inherently objective is incorrect because AI systems can reflect biases in training data and usage context. You need deliberate testing and oversight to prevent discriminatory outcomes.

Full AWS Practitioner Certification Question