Legal models hallucinate in 1 out of 6 (or more) benchmarking queries (hai.stanford.edu)