How would they even use the CSAM Detection System to root out terrorism?
It would only work if terrorist shared meme photos and they would soon learn not do that.
The system can't be used to detect a category of photos only a specific set of photos and close derivatives.
Indeed. But the set of photos is not disclosed, only their hashes. Therefore, it could contain any arbitrary photo, including photos matching known terrorists.
If the terrorist doesn't have a photo of himself in his phone, which would be reasonable to assume, then maybe he happens to be outside and someone happens to take a photo of him while taking a selfie or photographing a friend or any other activity people usually do outside, which is not at all unlikely.
And while the software as it is right now would be incapable of matching the terrorist's face at any angle, actual photo recognition could well be part of the next upgrade. And if we're adding photo recognition, might as well add GPS tracking too, so we know where the terrorist was when his face was captured. None of which would be unlikely, if they want to expand the idea of CSAM detection to detect terrorists in the first place.
It's not necessarily about the software in its current state (even though I could instantly see ways to circumvent and misuse it when it was announced, and I'm not even a security specialist), but the infinite possibilities for expansion.