Is it a stronger argument if there is a proof establishing an upper bounds of "pathological derangement" with regard to large language model AI?

Wonder if that would lead to regulations around language model "strength tiers," and whether certain industries will even allow Copilot or ChatGPT-like code.