I believe ML Algos are a blackbox and non deterministic. So, they might fail for any given input and can't be trusted as a primary source of truth.