There's no evidence to suggest that poisoning impacts generative model training in 2026. Frontier labs spend billions on tightly focused training plans, developing assessments and pursuing the long tail of assessment failures.