This paper discusses the importance of building guardrails for Large Language Models (LLMs) to mitigate their risks, particularly in the context of ethical use, data biases, privacy, and robustness. Guardrails are algorithms that filter the inputs and outputs of LLMs to ensure they do not generate harmful content. The paper reviews existing open-source solutions such as Llama Guard, Nvidia NeMo, and Guardrails AI, highlighting their strengths and limitations. It emphasizes the need for a systematic approach to building guardrails, involving a multi-disciplinary team to address complex requirements and conflicts. The paper also explores technical challenges in implementing individual requirements, such as vulnerability detection, protection via LLMs enhancement, and I/O engineering. It advocates for a neural-symbolic approach that combines learning and symbolic methods to handle complex cases and ensure robustness. The paper concludes by emphasizing the importance of rigorous verification and testing to ensure the quality and reliability of guardrails.This paper discusses the importance of building guardrails for Large Language Models (LLMs) to mitigate their risks, particularly in the context of ethical use, data biases, privacy, and robustness. Guardrails are algorithms that filter the inputs and outputs of LLMs to ensure they do not generate harmful content. The paper reviews existing open-source solutions such as Llama Guard, Nvidia NeMo, and Guardrails AI, highlighting their strengths and limitations. It emphasizes the need for a systematic approach to building guardrails, involving a multi-disciplinary team to address complex requirements and conflicts. The paper also explores technical challenges in implementing individual requirements, such as vulnerability detection, protection via LLMs enhancement, and I/O engineering. It advocates for a neural-symbolic approach that combines learning and symbolic methods to handle complex cases and ensure robustness. The paper concludes by emphasizing the importance of rigorous verification and testing to ensure the quality and reliability of guardrails.