Running model inference in trusted environments to protect proprietary models and sensitive query data.