ITERALIGN proposes a data-driven constitution discovery and self-alignment framework for Large Language Models (LLMs) to improve alignment with human values.