Of course. That's why it's important to reduce the attack surface. Adding electronics (or worse, software) adds a huge amount of attack surface. The attack could be at any point from the CPU-internals to the software.
> the voting protocol doesn't provide a means of verification
Yes. That's a feature. Any new system cannot re-enable voter coercion.
> Homomorphic encryption
I already mentioned[1] that video yesterday. It's an interesting idea, but even Prof. Rivest in the video isn't claiming it's ready for use.
More importantly, the reply by marten-de-vries[2] brings up a very good counter argument to any voting system based on fancy math: the general population won't accept it. The voting process doesn't work unless the population considers it legitimate, and it will be hard to convince them if they first have to learn enough math to understand homomorphic (or public-key) encryption.
This is still interesting research that may evolve into a new type of voting protocol in the future.