AI safety in the context of the frontier labs is completely a joke, much like privacy to the previous iteration of big tech. They talk about safety but end up just releasing the next more powerful model. They only talk about “scaling back” when the models aren’t getting better to protect their valuations.