Who Checks the Citations? Benchmarking Legal Hallucination Detection

This research tested AI systems designed to detect errors in legal citations using a specialized dataset. The study found that while advanced models like GPT-5 show high performance, they still struggle with subtle citation mistakes, especially content misrepresentation. The findings suggest that improving data access and targeted AI training are necessary because the problem of unreliable citations is worsening.

Open the full episode

More from AI Radio