This is not a dichotomy between perfection and zero. The efforts to restrict access to nuclear weapons have been very successful, even without being perfect.

Efforts to restrict large unaligned AI models may similarly buy us more years of existing.