Its a protection for the person who deploys the model and controls the context against some exploits, on the expectation that that operator will take steps to prevent large scale context manipulation.
It is not a protection for society against the person deploying the model.
It is not a protection for society against the person deploying the model.