Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems