Why not use the phone to detect physical and verbal abuse? The microphone is always on for most users, we have algos that can detect the sound of an adult yelling, a child crying, and the percussive sound of a human hitting another human. If 2/3 of those sounds are detected, why not forward that to NCMEC for further review? It doesn't have to be "1 yell, 1 cry, break down the door", it could be a pattern detector that logs the times, the severity, and the mode. If it hears objects being thrown, people yelling, kids crying, why isn't that sufficient to call authorities? How come "think of the children" suddenly doesn't work anymore?
I'm being 50% facetious here because obviously that's a huge privacy issue. But the same logic for CSAM scanning still applies nonetheless. Why only address 1/3 of abuse and ignore verbal/emotional, and physical? Those kids need justice too.