Term explained
What are AI guardrails?
Guardrails are the rules and filters that stop an AI system from saying or doing harmful, off-topic or embarrassing things.
In plain English
An AI model on its own will try to respond to almost anything. Guardrails are the safety layers added around it. Some are trained into the model itself, so it refuses to help with things like making weapons. Others sit around it: instructions in the system prompt (“only answer questions about our products”), and separate filters that check what goes in and what comes out.
Think of them like the barriers along a mountain road: they don’t drive the car, but they stop it going over the edge.
Why it matters
Guardrails are what make it safe for a company to put a chatbot in front of millions of customers. They block harmful content, keep the bot on topic, stop it leaking private information and prevent it from making promises the company can’t keep. When they fail, the results are often public and embarrassing.
An example
In 2024 a customer got the parcel company DPD’s chatbot to swear and write a poem about how bad the company was. After a system update its guardrails had stopped working, so it simply followed his playful instructions. DPD switched off the AI part to fix it.
What to watch out for
No guardrail is perfect. People keep finding ways around them, through prompt injection and clever wording known as “jailbreaks”. Guardrails can also be too strict and refuse harmless requests. If you run a chatbot, test it hard, and always give customers a way to reach a person.