Guard rails in AI are safety measures designed to prevent harmful or unintended outcomes. They ensure that AI systems operate within acceptable boundaries and adhere to ethical guidelines.