Guide

How to implement LLM guardrails

OpenAI Cookbook · developers.openai.com

Runnable input/output guardrail patterns, including the accuracy-latency-cost tradeoffs.

Open on developers.openai.com ↗

The question it answers

  1. How do you stop an AI support agent from hallucinating and breaking customer trust?

Read with it

  1. The landmark ruling every founder should know: you are legally liable for what your bot invents. forbes.com Article What Air Canada Lost In 'Remarkable' Lying AI Chatbot Case Forbes (Marisa Garcia)
  2. The Cursor incident as a business story: one hallucinated policy, hours of silence, real cancellations. fortune.com Article A customer support AI went rogue, and it's a warning for every company Fortune
  3. The technical blow-by-blow, including the fix Cursor shipped: labelling AI responses as AI. theregister.com Article Cursor AI support bot hallucinated its own company policy The Register (Thomas Claburn)
  4. The neutral, sourced record of the incident, good for internal postmortem-style learning. incidentdatabase.ai Article AI Incident 1039: Anysphere AI Support Bot Invents Login Policy AI Incident Database
  5. A concrete notebook for building a hallucination checker in front of customer-facing answers. developers.openai.com Guide Developing Hallucination Guardrails OpenAI Cookbook
  6. The most-cited engineering guide on keeping agents simple, bounded, and predictable. anthropic.com Guide Building Effective AI Agents Anthropic
eChai Partner Brands The partnership →