Veracity

Veracity is the responsible-AI dimension measuring whether a model’s outputs are truthful and grounded in evidence rather than merely fluent. Hallucination is its primary failure mode, and AWS addresses veracity through techniques such as Retrieval Augmented Generation (RAG) and Amazon Bedrock Guardrails, which anchor responses to approved knowledge sources and filter ungrounded content. The key exam distinction is between veracity and reliability: reliability measures whether a system performs consistently over time, while veracity measures whether individual outputs are actually correct. A model can be perfectly reliable — returning the same wrong answer every time — yet still fail on veracity.