Pre-screening would be the kind way of implementing this feature. They could make it so if CSAM is detected, just don't upload those files to the cloud and delete them locally. Then they are still doing due-diligence to prevent that material from entering their cloud and to remove that material from circulation, but they don't get random people arrested for false positives. When they detect hash matches on a device, they are under zero legal obligation to do anything about it because it could easily be a false positive (as the public has demonstrated on HN and elsewhere), so they'd be in the clear with this approach. This will have the effect they desire without pissing tons of people off.
Instead they have taken the aggressive stance of manual review + tipping off law enforcement, which goes against their entire mantra of protecting privacy. Your phone will now be an informant against you. If you are an activist or a whistleblower, a false positive could be enough to get your device searched and seized.