M&T Bank has placed AI copilots in the hands of more than 15,000 employees, extending the technology from drafting and summarization into customer service, software development and risk work.
The regional US bank’s rollout did not begin with unrestricted access. It initially blocked public large language models because employees might enter sensitive information, then tested enterprise alternatives with about 800 people before expanding the chosen system. That sequence places governance and internal technology modernization at the center of the deployment.
Copilots move into daily operations
Employees use AI to draft reports, analyze call-center conversations, generate code, identify customer needs and flag portfolio risks. M&T is also examining agentic systems for cybersecurity and fraud detection. A 2025 account cited by AI News said 16,000 of the bank’s roughly 22,000 employees were already using Microsoft Copilot for tasks such as email drafting and call summaries.
Chief data officer Andrew Foster said summarizing a call-center conversation with generative AI saves about six minutes per call. Developers also use GitLab tools to produce code, but employees remain responsible for reviewing generated work.
That obligation appears in M&T’s 2026 business conduct policy. Staff must use approved tools and may not enter confidential, proprietary, customer, employee or regulated information into unapproved systems. The policy leaves accuracy and appropriateness with the human employee rather than treating an AI output as an approved final answer.
Modernization created the operating base
The AI expansion follows a technology overhaul that began in 2018. At that time, the bank says more than half of its technology specialists were external workers. It now reports that 80 percent of the technology workforce is in-house, with about 2,000 technologists distributed across more than 300 agile teams and over 1,000 specialists hired during the program.
M&T also replaced dozens of legacy platforms. The bank says technology outages have fallen by more than 80 percent since 2018, while annual system upgrades increased by 300 percent. Technology spending exceeded $1.2 billion in 2025, almost three times the 2017 level, and annual releases reportedly rose from about 15,000 in 2018 to 65,000 in 2025.
Those performance figures come from M&T executives and reporting cited by AI News; the source does not provide an independent audit. They nevertheless show that the copilot rollout sits inside a broader attempt to change staffing, release frequency and infrastructure rather than operating as a standalone software purchase.
Governed data supports three AI routes
Foster began building a data-lineage program after joining the bank in 2023. He described lineage as a general capability for understanding where information originates, how it moves and how it is used, not a project created solely for generative AI.
M&T now uses an internal repository called Edison for authoritative policy documents, along with Solidatus and Monte Carlo software to trace information through databases, applications and business-intelligence systems. Around 2,000 employees have participated in a Data Academy focused on governance and skills. The bank also uses retrieval-augmented generation with governed internal data.
Senior technology executive Mike Wisler described three routes for adoption: general employee copilots, AI functions already embedded in existing applications and proprietary systems built around M&T’s data and processes. With more than 1,800 applications, many supplied by outside vendors, the embedded route alone requires the bank to decide which capabilities are useful and acceptable.
The source documents broad adoption and several claimed efficiency gains, but it does not quantify the total return on the bank’s AI spending or the accuracy of individual systems. M&T’s approach is therefore best understood as a controlled expansion: employees gain more automation, while approved tools, human review and traceable internal data define the boundary around its use.
