Researchers introduce DAVinCI framework to make AI language models cite sources and verify facts, reducing hallucinations in healthcare and legal work

A team of AI researchers released DAVinCI, a new verification system designed to reduce false claims from Large Language Models (AI systems that generate text). Unlike standard AI, DAVinCI traces each claim back to its source—either information learned during training or external documents—then checks if that claim is actually supported by the evidence, similar to fact-checking journalism.

The system works in two stages: it first identifies which parts of the AI's reasoning came from which sources, then validates each claim using logical reasoning (checking whether conclusions follow from stated facts) and confidence scoring. Testing on fact-checking datasets like FEVER shows DAVinCI significantly improves accuracy compared to AI systems with no verification.

For professionals in healthcare, law, and finance who rely on AI summaries or research assistance, this matters because sources become traceable and verifiable—you can see exactly where the AI pulled information from and whether it's actually correct. This reduces the risk of AI confidently stating false information (called 'hallucinations'), which could otherwise lead to wrong diagnoses, flawed legal arguments, or bad business decisions.