As customer-facing GenAI support systems move from prototypes to production, reliability is no longer a nice-to-have—it’s a hard requirement. Just ask the teams behind NYC’s MyCity chatbot, which advised businesses to break the law; Air Canada’s support AI refund assistant, which hallucinated non-existent policies and cost the airline real money; or Cursor’s AI-powered support agent, which confidently fabricated a login policy and triggered a wave of user cancellations. These public failures aren’t edge cases. They’re symptoms of a deeper reliability gap in how we build and monitor AI-driven support.

In this session, we’ll explore common failure modes in customer-support GenAI systems—from factual mistakes and hallucinations to deeper reasoning and alignment issues—and walk through practical strategies for detecting and remediating them. We’ll demo how to integrate real-time trustworthiness checks into Retrieval-Augmented Generation (RAG) and Agentic systems, and show how Cleanlab AI platform can be used to deliver expert-verified, zero-hallucination responses; enforce guardrails for safety, compliance, and helpfulness; surface root-cause insights to diagnose retrieval failures and documentation gaps; and route customer intent to the right support channel. We’ll also discuss how to incorporate human-AI collaboration to strengthen accountability and trust.

If you’re building GenAI customer-support agents that users or businesses rely on, this talk will help you close the gap between promising prototypes and production systems you can stand behind.