Talent depth and research focus reveal which banks are building scalable AI capability

AI research maturity moves beyond publications

Across 2872 papers between 2023-2025, the top five banks produced 56% of AI research papers, but publication volume is becoming a less complete measure of AI maturity as leading institutions deepen proprietary AI development. 

For challengers such as Truist, NatWest and Santander, publication growth still matters, showing where banks are formalising AI research through labs, senior leadership, and structured teams.

AI research talent is concentrating at the top, but challengers are closing in

Publication leaders also hold the talent advantage. JPMorganChase, TD Bank, Capital One and RBC lead the sector on AI research talent, giving them the scale and role mix needed to turn research into capability. GenAI is raising the bar further, creating demand not only for frontier research expertise, but for engineering and applied skills that make AI systems work in practice. 

Other banks may struggle to match this scale, but NatWest, Morgan Stanley and Santander are building focused capability through selective hiring in specialist and research engineering roles.

Agentic AI moves toward production, but cost emerges as the next hurdle

Agentic AI now accounts for 11% of banks’ AI research, with work focused on making agents reliable, governable and useful in complex workflows. JPMorganChase has deployed a customer-query annotation system that cleared a 1 million-case backlog and saved over 5,000 annotation hours a year, while Capital One is applying multi-agent systems to cybersecurity incident response and model risk and compliance audit. 

Cost is the next constraint. As agentic systems start to be deployed in real-world settings, banks will need to manage not only whether they work, but whether they can run efficiently and economically at scale. GenAI inference optimization is already a small but fast-growing research area, with its share of banking AI research doubling in 2025 as institutions look to make proprietary GenAI cheaper to run at scale.