Guardrails
Also known as: Safety filters, Content moderation
Pre- and post-processing layers that block bad inputs/outputs and enforce policy on top of an LLM.
Input/output filters and policies layered around an LLM to block disallowed content, prevent prompt injection, enforce output schemas, or gate sensitive actions. Implemented via classifiers, regex, smaller LLMs, or rule engines.
In practice
Every production LLM app needs them. Interviews probe layered defense, latency budget, and false-positive trade-offs.
Related topics
Questions that mention this term
Related terms
Chatbot Arena
Crowdsourced blind A/B comparisons of LLM responses, aggregated into an Elo leaderboard.
Eval Harness
A framework for running, scoring, and tracking LLM evals over time; handles golden sets, judges, and regressions.
Golden Set
Curated test cases with expected outputs; your regression suite for prompt and model changes.
Hallucination
When a model confidently makes up something that isn't true.
HumanEval
Code-generation benchmark: 164 problems with hidden unit tests, scored by whether the generated code passes.
Jailbreak
Adversarial prompt that bypasses an LLM's safety training to elicit normally-refused content.