Risk & Hallucination Rates

As a directional estimate, error and hallucination risk varies widely by task type rather than applying as a flat rate across legal AI use. Document summarization sits at the low end (1.8%), while tasks requiring precise sourcing or cross-jurisdictional reasoning score highest: litigation strategy at 8.5%, jurisdiction analysis at 7.3%, and case law citation at 6.2%. The pattern suggests risk concentrates where a model must anchor claims to specific external authorities rather than restate or synthesize provided text, reinforcing the case for citation-verification layers on higher-stakes research and drafting workflows.