AI & ML · Jun 15
Air Canada's Chatbot Lost in Court. The Model Was Fine. The Governance Was Not.
Five real-world AI failures show that when deployments go wrong, the culprit is almost never the model itself.
- AI production failures are almost always governance failures: missing review layers, unclear scope, and no human oversight in the loop let normal model errors escalate into legal and financial losses.
- Before deploying any customer-facing AI, explicitly define what the system can commit to, who reviews high-stakes outputs, and what happens when the model is wrong.
- Corporate AI research is increasingly focused on pre-deployment work, leaving deployment-stage risks like hallucinations and bias under-studied, so practitioners cannot wait for academia to solve this for them.