It's because these industries don't create their own benchmarks. The only ones creating evals are the AI company themselves or open source software engineers