Guardrails
Safety rules programmed into an AI to prevent it from generating harmful, illegal, or highly offensive content.
In Plain English
Guardrails are the digital bumpers on a bowling alley. AI companies install them to keep the AI's behavior within safe, acceptable limits. If you ask an AI for instructions on how to build a bomb, the guardrails trigger a canned response like, "I cannot fulfill this request." While necessary for safety, they can sometimes be too strict, refusing to answer harmless questions.
Real-World Example
An AI refusing to write an essay containing hate speech because it violates the company's safety guardrails.