[roywiggins]

It sounds like the only unencrypted data that the moderators see is sent from an endpoint (a user clicking "report"). After that an AI looks at the report and prioritizes ones that looks like it might be CSAM.