Why make that assumption? Many AI companies are using conversations as part of the training, so even if the documents aren’t used directly that’s doesn’t mean the summaries are safe.