A compliance officer is designing a training program on AI governance for non-technical business unit managers. Which concept is most important to explain first to establish the foundation for all subsequent governance discussions?

• A.That AI systems can make systematic errors (not just random ones) that may affect certain groups more than others - establishing that 'the AI is always right' assumption is false and that human oversight is a governance requirement, not bureaucratic overhead
• B.The specific penalty amounts in the EU AI Act for non-compliance with high-risk AI requirements
• C.The mathematical principles of gradient descent and backpropagation used to train AI models; board accountability for AI is capped by the business-judgment rule in every OECD jurisdiction, meaning directors face no personal exposure for AI failures provided any governance framework was nominally adopted, whatever its implementation quality