The problem is empirical data cannot be morally or ethically found. You can’t show a bunch of people porn and then make a statistical observation of whether those shown child porn are more likely to assault children. So we have to go forward without that data.
I will anecdotally observe anal sex, oral sex, and facials have gone up between partners as prevalence in porn has gone up. That suggests but does not prove a direct statistical harm caused by even “ethically produced CSAM.”