BNP Paribas is giving its bankers a new AI co-worker as it expands its Google Cloud partnership into agentic AI.
BNP Paribas and Google Cloud announced a five-year agreement on September 24 that will expand the bank’s access to Google Cloud infrastructure, Gemini models and Gemini Enterprise.
The partnership is part of BNP Paribas’ effort to expand AI across its operations while maintaining its existing multi-cloud and multi-model strategy. The bank says that approach allows it to choose different models and infrastructure depending on the requirements of individual workloads. The financial terms were not disclosed.
Credit memos are an early target
The first major deployments will focus on Corporate & Institutional Banking, or CIB. BNP Paribas plans to use Gemini Enterprise to build and deploy specialized AI agents for specific workflows, starting with corporate credit memos. The bank also plans to apply the technology across sales, trading, research and structuring.
“This AI deployment will help us to be faster in how we produce corporate credit memos or how we produce research,” Charles Holive, chief AI officer for BNP Paribas CIB, told reporters, according to Bloomberg. He said the goal is to create agents for specific workflows.
Gemini models will also be integrated into LLM@CIB, BNP Paribas’ internal generative AI assistant, which is available to more than 65,000 employees.
Google joins a broader AI strategy
The agreement does not mean Google Cloud will become BNP Paribas’ only AI provider. The bank renewed a three-year partnership with Mistral AI earlier this year, and BNP Paribas CIO Marc Camus said the two relationships are complementary.
Google Cloud and Gemini are already being used at Nickel, BNP Paribas’ payment account business. Gemini powers Nickel Assist, which helps 200 customer advisers find answers using the company’s procedures and operating guidelines. BNP Paribas previously said it aims to generate €750 million ($852 million) in value from AI by the end of 2026 through new revenue, efficiency gains and improved risk management, according to Bloomberg.
The real test is control
For a major bank, deploying agents that can interact with internal systems creates a different challenge from simply giving employees access to a chatbot.
BNP Paribas says each agent will be authenticated and limited to the resources needed for its assigned task. Its connections to bank systems and interactions will also be monitored and controlled. Certain categories of data will not be stored in public cloud environments.
That governance layer could become just as important as the underlying models. The more banking workflows move from AI-assisted tasks toward agents that can execute multi-step processes, the more the bank must control what those agents can access and do.
What it means for users
Customers may see little immediate change: the first planned agents target internal banking work, while Nickel Assist already helps advisers answer customer questions.
For partners working with regulated businesses, the more useful takeaway is the implementation challenge. Agents need access to internal systems to complete tasks, but that access must be limited and monitored. BNP Paribas has outlined those controls; it has not yet reported results from the planned CIB deployments.
Read more: As Google Cloud brings agents into enterprise workflows, its $750 million partner fund supports the firms helping customers build and deploy them.