During a pre-implementation risk assessment, an AI model is determined to present a significant risk...

The AAIA™ Study Guide advises that if an AI model presents a risk that exceeds the organization's predefined risk tolerance—especially in cases of ethical harm or bias—deployment should be delayed until proper safeguards are in place. This approach prevents legal exposure and preserves stakeholder trust.

“When AI risks exceed acceptable thresholds, organizations must suspend implementation until corrective action reduces the risk to within tolerance levels. Proceeding without mitigation violates sound governance principles.”

While improving data (B) may help, it does not address the immediate governance concern. Risk tolerance (D) should not be adjusted to fit flawed systems. Thus, A is the correct course.