Itaú Unibanco, Brazil's largest private bank with over 100 million customers, has deployed two distinct generative AI systems: ia.i, a customer-facing conversational interface, and IARA, an internal governance platform for employees. Both run on OpenAI models. The bank operates under Brazil's Central Bank regulations, one of Latin America's strictest financial compliance environments, making this deployment technically and legally significant.
The architecture choices matter more than the headline. IARA is not a chatbot. It is a governance layer that routes AI requests across business units, enforces policy, and logs decisions for regulatory audit. ia.i handles customer queries in Portuguese with context retention across sessions. The full article details how Itaú structured model access controls, handled data residency requirements, and measured accuracy thresholds before production rollout.
Itaú's case is worth reading because it is a stress test, not a pilot. Scaling generative AI to 100 million users in a regulated sector with a parallel internal governance system is a template other large financial institutions will copy or reject. The specific failure modes, guardrails, and compliance tradeoffs Itaú documents are the actual value here.
[READ ORIGINAL →]