How does BNY Mellon use Gemini in Eliza?

BNY is turning its internal AI system into a test case for how agentic models function in a regulated financial institution. The bank said it will integrate Google Cloud’s Gemini Enterprise into Eliza, the AI platform it launched in 2024. The announcement came on December 8.

Eliza sits at the center of BNY’s effort to embed AI across the firm. The bank has more than $57.8 trillion in assets under custody or administration and $2.1 trillion under management.

BNY launched Eliza as a model-agnostic operating system that allows employees to build and deploy AI agents. Adoption has been rapid. A July report from PYMNTS said 96% of employees use the platform.

The Gemini Enterprise integration adds multimodal capabilities for document analysis, research support, and workflow automation. Google Cloud said the system is designed to help employees process financial data and generate insights across business lines.

BNY’s chief data and AI officer, Sarthak Pattanaik, said the move advances a strategy of “AI for everyone, everywhere and everything.”

The bank has positioned this as capacity expansion, not headcount reduction. CEO Robin Vince told Time in June that most employees had completed generative-AI and responsible-AI training.

Scaling an AI Workforce

BNY operates in a domain where errors carry regulatory consequences. Much of its business involves custody, asset servicing, and post-trade operations. These processes depend on document validation, compliance checks, payment-instruction verification, and reconciliation. The bank has already deployed more than 100 AI solutions.

Agentic systems are designed to handle multi-step tasks rather than generate text for a single prompt. The models support reasoning across documents, structured data, and historical inputs. According to reports, BNY employees will be able to build agents that analyze financial reports and automate research tasks.

BNY’s AI architecture differs from the approach taken by peers. JPMorgan distributes access to several external large-language models but does not use a unified internal system.

Goldman Sachs rolled out a firmwide assistant in June to help summarize documents and support engineering workflows. Citigroup and Wells Fargo have expanded pilots across compliance and operations but continue to rely on distributed tooling.

BNY has instead built an internal platform that centralizes model governance, agent deployment, and enterprise controls. The bank publishes its AI ethics framework, which covers model review, bias testing, explainability, and access restrictions.

The Risk Controls Question

The scale of this integration raises questions for supervisors. The OECD warns that concentration of AI workloads in a small set of cloud providers could create systemic vulnerabilities.

Regulators have also pointed to the difficulty of validating AI models that perform multi-step reasoning. The Financial Stability Board said in October that AI adoption increases model-risk complexity and can obscure decision paths.

BNY said Eliza’s design keeps humans in the loop and maintains full audit trails. Its “digital worker” agents are required to log actions and operate within permissioned systems. The bank has described this architecture in public forums but has not released technical details.

Financial institutions have accelerated AI deployment in the past two years. An EY-Parthenon survey found that more than three-quarters of banks had launched generative-AI applications.

Consultants and industry analysts expect productivity gains but caution that converting these gains into revenue remains difficult. Reuters reported last year that banks have struggled to turn AI adoption into measurable profit despite operational improvements.

BNY’s scale, combined with its custody franchise, gives it access to transaction data and documentation from across the financial system. This provides a large operational surface for AI deployment but also subjects the bank to heightened scrutiny.

BNY is treating AI as core infrastructure rather than an experimental toolset. The bank’s partnership with Google Cloud brings agentic capabilities into an institution that operates under global regulatory regimes. The key indicators for the success of this initiative are governance, accuracy, and the ability to maintain controls as agents take on more complex workflows.

The agreement positions BNY as one of the first major banks to test agentic AI inside core financial operations at scale.