# What are guardrails in AI? — QANATONE > Guardrails are the rules and filters that check what goes into and comes out of an AI system, keeping it on topic and preventing harm. Kaynak: https://www.qanatone.com/en/nedir/guardrail-nedir/ --- [Explainers](https://www.qanatone.com/en/nedir/) # What are guardrails in AI? 15 Sep 2026 2 min read Safety · Guardrails Guardrails are the rules and filters that check what goes into and comes out of an AI system, keeping it on topic and preventing harm. ## What are guardrails? Barriers on a motorway keep cars from leaving the road. In AI, guardrails do the same job: they are checks that limit what a model accepts and what it returns. Models come with safety behavior from their own training, but that is not enough for a product; the company building it needs an extra layer based on its own rules. An example of why: in 2024 a Canadian airline was held liable by a tribunal for a refund policy its chatbot had described incorrectly to a customer. ## What gets checked? - Input checks: messages are scanned for attempts to trick the model (prompt injection, jailbreaks), personal data such as ID and card numbers is masked, and off topic requests are filtered. - Output checks: answers are checked for harmful, abusive or legally risky content, facts such as prices and stock are compared with real system data, and the output format is validated. - Action checks: what an [agent](https://www.qanatone.com/en/nedir/ai-agent-nedir/) may do is restricted, and steps such as payments and deletions require human approval. ## How are they implemented? Guardrails are built in layers. The simplest layer is rules in the [system prompt](https://www.qanatone.com/en/nedir/system-prompt-nedir/), but on its own it is not reliable. Stronger methods are separate checks placed before and after the model call: word and pattern filters, classifier models (such as Meta's Llama Guard) or open source tools like NVIDIA's NeMo Guardrails. For critical decisions the last layer is a person. ## Getting the balance right Guardrails limit the damage from [hallucination](https://www.qanatone.com/en/nedir/halusinasyon-nedir/) and misuse. The dose matters too: an overly strict set stops the model from answering harmless questions and drives users away. A good set is tuned against real conversations and reviewed regularly. ## More articles [Reliability · Hallucination ### What is AI hallucination and why does it happen? 15 Sep 2026 · 2 min](https://www.qanatone.com/en/nedir/halusinasyon-nedir/)[Infrastructure · Inference ### What is inference in AI? 15 Sep 2026 · 2 min](https://www.qanatone.com/en/nedir/inference-nedir/)[Marketing · Influencers ### What is influencer marketing and how does it work? 15 Sep 2026 · 2 min](https://www.qanatone.com/en/nedir/influencer-marketing-nedir/)[Performance · Latency ### What is latency in AI applications? 15 Sep 2026 · 2 min](https://www.qanatone.com/en/nedir/latency-nedir/)[Model Type · LCM ### What is a Large Concept Model (LCM)? 15 Sep 2026 · 2 min](https://www.qanatone.com/en/nedir/lcm-nedir/)[Model Type · LAM ### What is a Large Action Model (LAM)? 15 Sep 2026 · 2 min](https://www.qanatone.com/en/nedir/lam-nedir/)[Marketing · Basics ### What does marketing mean? Definition and the 4Ps 15 Sep 2026 · 2 min](https://www.qanatone.com/en/nedir/marketing-ne-demek/)[Model Type · MLM ### What is a masked language model (MLM)? 15 Sep 2026 · 2 min](https://www.qanatone.com/en/nedir/mlm-nedir/)