Risks and safetyEasy read
Guardrails
Written by Minh Khang
What is Guardrails?
Guardrails are hard rules around AI so it does not say or do things outside its remit, such as giving a bigger discount than allowed.
Guardrails can be instructions in the prompt, a filter that checks answers before they go out, or limits built into the tools (no refund above a set amount). Rules in software hold better than instructions, because AI can still ignore an instruction.
Guardrails in a shop
The bot may not offer more than 10% off: any reply with a bigger discount is held for a person to approve.
Keep learning beyond the glossary
Read next
- Agents and automationHuman in the loopHuman in the loop is a design where a person reviews and approves before AI does something important: the AI drafts, the person decides.
- Risks and safetyPrompt injectionPrompt injection is when someone hides instructions in a message or document to trick an AI into doing what they want.