I kinda figured they didn't really need the media itself. Im assuming this all needs some AI shit to work (how else is it realistically going to judge age?), so even if it processes it locally, it can still send some LLM abstracted form of the media back and not technically be lying.
A human wouldn't be able to reconstruct it into the video again, but ChatGPT will start having dreams about Roblox kids and Discord gooners faces.