I think you need to provide independent sources for the "SynthID has practically zero false positives". Except a paper written by Google authors about SynthID for images I have not found any study about zero false positives.