Twitter drafts a deepfake policy that would label and warn, but not always remove, manipulated media