This is covered in the paper. They acknowledge this defense technique while also pointing out that a laser flashlight could be used to illuminate all microphones at once.
But in some future rev, one could imagine that if the mics in the array are non-coplanar (e.g. at least 4 mics) and sufficiently far from each other, then there is no possible way for the audio signal to reach them at once (unless it is actually light being measured).
However, non-coplanar mics would work for the opposite reason: If they are on different sides of the device, you couldn't reach all of them from the same distant location. So unless all mics receive (more or less) similar sound signals, you could discard it as manipulation.