Putting aside the obligation(s), the mere fact they (and to be fair, likely other clouds) have an automated deletion process for unwanted content reduces their reliability. There may be bugs, possibly poor hash distributions, malicious misreporting, erroneous classifications, or misappropriation of the tool for other purposes.

I don't think this measure is effective (it's trivial even for non-techies to circumvent hash-based detection) nor proportional to the likely incidence/impact of this crime. While I cannot speak to what their legal obligations are, I don't think we should encourage invasion of privacy for no particularly good reason. As such, I strongly object to the notion they have a moral obligation to commit this particular harm. Facetiously claiming that doing so somehow protects children is just a joke in particularly bad taste.