I feel like there was a meeting about data security risks at the White House, with CEOs from Open AI and Google, as well as a doc, calling out this exact situation (recording of metadata, queries). Is there really no consistent way of filtering out classified/restricted information from Open AI's GKR?