There has to be a way to connect back to it for the model to be able to take input.
Its can easily be a pull-only so inference server the only one that might initiate connections.
In any case if you know how inference works there basically nothing you can exploit in tokens processing, it's just basic math.
After all it can be just some GPU server spewing text over network. Like there are no way to connect back to it.
Nothing except inference running on servers with GPU so there just nothing to "hack".