But if that's true for attacking, I don't see how it's not also true for defence? If LLMs make it easier to attack, and this is a generally available model, the

But if that's true for attacking, I don't see how it's not also true for defence? If LLMs make it easier to attack, and this is a generally available model, then it can as well be used for defence, and easier too, since you have the source too.