ML Agents: Why does heuristic control ignore masks?

From the documentation:

When using Discrete Actions, it is possible to specify that some actions are impossible for the next decision. When the Agent is controlled by a neural network, the Agent will be unable to perform the specified action. Note that when the Agent is controlled by its Heuristic, the Agent will still be able to decide to perform the masked action.

Why is this? Once it's off the neural network, is my AI suddenly allowed to make illegal actions? Do I have to rely on training the AI so well to the point that it would never make such actions, or something? I don't understand...