Guardrails
Technical and policy layers that constrain what an AI system may say or do: filters, rules, tool permissions, and human review.
Guardrails are the practical fences around AI behavior: input/output filters, blocked topics, allowed tools, rate limits, logging, and escalation to humans. They implement alignment goals in production, not just in training.
For LLMs, chatbots, and AI agents, guardrails sit beside good prompt design and testing. They reduce, but do not eliminate,jailbreaks, prompt injection, and harmful completions. Teams often validate them with structured red teaming.