How do you know an LLM answer is actually grounded — not just plausible? I measured it across 7 models and 4 regulated domains

Brian Barbour developed a sophisticated auditing system to verify if large language model (LLM) responses are accurately based on the source texts they claim to reference, testing it across seven different models and four regulated domains. The core issue is that LLMs can generate answers that sound correct but are actually fabricated or misleading. Barbour's method aims to uncover these discrepancies, which is crucial for ensuring the reliability of AI-generated information, especially in regulated fields. The results and methodology invite technical scrutiny to improve the verification process for LLM outputs.