There is a growing race on Wall Street to deploy AI agents that can handle end-to-end tasks.
BNY announced Monday that it will incorporate Google Cloud's agent AI technology, including Google's latest model Gemini 3, into the bank's internal AI platform Eliza, named after the wife of bank founder Alexander Hamilton. The technology stack upgrade is the latest step the bank, one of New York's oldest operating banks, is taking to advance its AI ambitions.
By integrating Google Cloud's technology into Eliza (which already leverages multiple large-scale language models), the company believes it will enable employees to handle daily tasks faster, Sarthak Pattanaik, the bank's chief data and AI officer, told Business Insider. Eliminate the tedious manual work of client onboarding. Staff collect documents, verify identification and tax forms, identify important details, search for risk information, and log everything into internal systems. He said agent AI could help orchestrate these steps, breaking down tasks into smaller components and providing needed information in a more streamlined flow.
“Eliza can now orchestrate processes more simply, efficiently and seamlessly through agent technology workflows,” said Pattanaik.
Gemini 3, which debuted in mid-November, can interpret text, images, tables, PDFs, and audio together, allowing employees to load mixed financial documents and let the model interpret and synthesize what's important.
BNY's generative AI build-out began to accelerate in 2023, and the bank was keen to position itself as an early adopter of AI. Eliza currently supports over 120 automation tasks. Pattanaik added that almost the entire company has completed generative AI and responsible AI training.
During BNY's third-quarter earnings call, BNY CEO Robin Vince told analysts that the bank leverages agent AI to deploy more than 100 “digital employees” who “work in tandem” with staff to perform tasks such as payment verification and code remediation.
“We believe the AI opportunity is significant and are pursuing it with urgency,” Vince said.
Earlier this year, we announced a similar partnership with OpenAI, maker of the popular large-scale consumer language model ChatGPT. The company's website describes it as “the first major bank to deploy an AI supercomputer (powered by NVIDIA) to accelerate processing power.”
safety protocols
While exciting, the extension to agent systems raises questions about how highly regulated institutions like BNY ensure data privacy. Both BNY and Google emphasized that deploying agent AI within regulated institutions requires delineating boundaries around what the technology can see, decide on, or escalate.
Pattanaik said the introduction of agent AI within banks requires strict oversight. He said each agent must pass an internal model risk review before it begins operations, and stressed that the system is governed by strict access controls that determine what information can be used.
Once deployed, the team will monitor agent performance daily and incorporate the results into an ongoing feedback loop, he added.
Rohit Bhatt, Google Cloud's head of financial services, told Business Insider that Google provides safety mechanisms that limit how agents communicate and what data each agent can access. Burt said agents have a “development kit” and “protocols” to manage their communication with each other.
“The main purpose of the kits and protocols is to establish communication channels with boundary conditions such as, 'You are only allowed to talk to this agent for certain reasons and nothing else,'” he added.
Why we think Google can help Wall Street
Large banks are employing a mix of homegrown and external AI tools. Goldman Sachs has expanded its internal platform while experimenting with technology from startups like Cognition Labs.
Dealmakers and investors are embracing tools from new entrants like Hebbia, which offer rapid libraries and deep search. Morgan Stanley is deploying OpenAI technology to help its financial advisors, and JPMorgan executives have spoken publicly about the possibility of junior employees managing teams of AI agents.
Bhatt said financial firms are becoming important testing grounds for agent AI because their workflows involve large amounts of documentation, strict rules, and significant risk management.
Google's pitch is that Gemini can reason through long material while remaining tightly compliant with a company's internal policies. This is a prerequisite for deploying agents across management, marketplace, or onboarding.
“We need to make sure that whatever these agents are doing is based on the context of the business and the specificities of the business,” he said. “To do this, these models need to be able to understand and adhere to specific policies and rules, ensuring they meet the ‘expected standards’.
