We measure how AI systems fail together, so correlated risk can be priced and insured instead of excluded.