Governance failures are usually failures of scope — regulators move too slowly, apply old frameworks to new harms, miss edge cases. South Africa's withdrawal is a different category: the institution responsible for AI oversight produced a document that demonstrates the core failure mode of AI deployment — confident, well-formatted output built on sources that do not exist . The problem is not that a government used AI to help draft a policy. The problem is that no verification layer caught the hallucinations before publication.