JPMorgan Leads as Five Banks Command 65% of Banking AI Research

Five banks dominate AI research in finance - here's what matters for your roadmap

As of September 2025, a new study shows five banks produce 65% of all AI research in the sector. JPMorgan Chase leads with 37% of publications, followed by Capital One (14%), and Wells Fargo, RBC, TD (5%, 5%, 4%).

The analysis reviewed 2,700+ AI-specific papers from 50 of the largest banks. Annual output has grown about 7x in five years, and the number of banks publishing since 2019 rose from 25 to 46.

Who's driving the agenda

• JPMorgan Chase: 37% of sector publications
• Capital One: 14%
• Wells Fargo: 5%
• RBC: 5%
• TD Bank: 4%

Research isn't sitting on shelves. RBC's ATOM model supports lending decisions, and Capital One's multi-agent systems are being applied to customer service. The signal is clear: papers are translating into production.

Agentic AI is moving up fast

Papers on AI agents and agent-based systems are now the 5th most popular theme, about 6% of 2025 YTD publications. That's roughly double the share of public agentic deployments seen so far, pointing to near-term pilots and launches.

If you're evaluating agent workflows, align your teams on definitions. A quick primer on multi-agent systems can help standardize language across Ops, Risk, and Tech.

Why this matters to finance leaders

The top banks are setting direction on what "good" looks like in a regulated setting. Expect a 2-3 year cycle from research to scaled production for the most investable ideas.

Evident's CEO, Alexandra Mousavizadeh, put it bluntly: "Through their research programmes, banks like JPMorganChase, Capital One, RBC, Wells Fargo, and TD Bank are setting the tone for how AI will be deployed in high-stakes, regulated environments... moving from research pipelines into production at scale within two to three years, which is lightning fast by academic standards."

Actions to take this quarter

• Set a research-to-production cadence: Map a 24-36 month pipeline with gates for data, compliance, and ROI. Fund it explicitly.
• Staff for applied science: Hire a small core of research engineers, then augment with partnerships and targeted contractors.
• Prioritize agentic pilots where guardrails are strong: Contact center triage, underwriting co-pilots, fraud casework, collections. Start with human-in-the-loop.
• Tighten model risk for agents: Role definitions, tool access limits, audit logs, escalation policies, adversarial testing, and clear fallback rules.
• Measure what matters: Cost-to-serve, handle time, first-contact resolution, approval speed, loss rates, complaint ratios, compliance exceptions.

Signals to watch

• New bank papers on agents, retrieval, and decision support
• Announcements of production launches that mirror prior research themes
• University collaborations and bank-hosted datasets or benchmarks
• Hiring spikes for research engineers and model risk specialists

Bottom line

AI research in banking is broadening, but direction sits with a handful of firms that can fund, test, and deploy at scale. Track their publications as a lead indicator for what will hit production in the next two years-and position your budgets, teams, and controls accordingly.

For a practical scan of tools already usable in finance, see AI tools for finance.