It’s not much of a deterrent to a motivated organization.

They are basically saying that models themselves are already capable. If you fine tune them with some cyber security tasks, that should make them potent. :-/