Inside Citi’s AI Playbook: Curated Data and Transparent Decisions

Jonathan Lofthouse from Citi talks about how the bank is making sure its use of AI is transparent and grounded in reliable data and explains that many of their AI tools are designed to show the underlying data they rely on, so users can see where an answer is coming from rather than treating it as a black box.

He gives the example of their “Citi assistant,” which is built on a specifically curated Citi-related data which doesn’t roam the open internet; instead, it uses trusted, public-domain information that Citi has selected for that purpose. The idea is to keep the model’s responses tightly linked to controlled, high-quality sources that are appropriate for the use case.

Lofthouse also notes that Citi does have some AI use cases that do draw on the wider internet. In those situations, the tools are set up to clearly cite external references, and to surface both the model’s reasoning and the data it used. For colleagues inside the bank, that transparency is crucial: it lets them check that the sources are credible, that the logic makes sense, and that the output stands up in the context of their work in financial services.