As long as we don't drug the hamsters and jam them into some sort of fake digital world, we should be fine? Put OpenAI's security researchers on it, heard they're good at containing experiments.