Hybrid AI governance has become essential for regulated industries such as banking, where the pressure to move fast with agentic AI collides with strict requirements for data sovereignty, compliance and model control.

For Europe’s largest bank, AI transformation is not a single cloud migration or a proof-of-concept sprint — it is a multi-year industrialization effort built around governed, deliberate use-case scaling. BNP Paribas SA now has nearly 1,000 of those AI use cases underway across its global operations, with a structured AI factory at the center of its approach, according to Jean-Michel Garcia (pictured), group chief technology officer at BNP Paribas.

“We’ve been working for models and algorithms for a while now, almost 20 years, so it’s not something new,” Garcia said. “[What] came as a surprise two years ago [was] the impact, the scale of the transformation and the fact that in one single new technology, we were able to maybe transform all kinds of business we used to have — all the employee experience, all the customer experience.”

Garcia spoke with theCUBE’s John Furrier at the Think 2026 event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed BNP Paribas’s hybrid AI governance strategy, AI factory implementation and balancing speed with data sovereignty. (* Disclosure below.)

Hybrid AI governance and the AI factory model

To avoid the mistakes of prior digital transformation waves, BNP Paribas took a deliberately structured approach from the start. Rather than consolidating all stakeholders into a single initiative — a move leading to “a big, big mess” in the cloud era — the bank established a group-level AI governance framework with federated governance by business and function, connected through an internal AI factory and an LLM-as-a-service platform, according to Garcia.

“The strategy is to use AI, not to become AI — to use AI to better serve our customers and to give more value to them,” Garcia said. “But also internally, for employees, to give them more capabilities to focus on the right set of tasks and to make them more efficient and more focused on what they do.”

The coding assistant journey illustrates the broader evolution. Early developer adoption relied on raw LLMs outside established tool chains, but over the past year the practice has spread across the organization as tooling matured and business value became visible, Garcia noted. The bank is now evaluating multiple coding platforms simultaneously — including GitHub Copilot and Cursor — partly as a deliberate hedge against vendor lock-in as the model landscape continues to shift rapidly.

“Don’t play with your data. Don’t underestimate the risk of spreading everything outside,” Garcia said. “Even [when] you need to do things with dedicated partnerships outside the bank, clients [are the] most important.”

Stay tuned for the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the Think 2026 event.

(* Disclosure: TheCUBE is a paid media partner for the Think 2026 event. Neither IBM, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE