Financial crime compliance was not the obvious starting point. Yet when FIS, one of banking’s most consequential infrastructure providers, chose where to deploy its first agentic AI product built with Anthropic, it chose anti-money laundering investigations. That choice carries a logic worth understanding, because it maps almost exactly onto the pattern that defines every durable technology transition in banking: start where the risk of getting it wrong is high enough to demand rigor, build the governance muscle, then expand.

The announcement, made on May 4, 2026, confirmed that BMO and Amalgamated Bank are among the first institutions to run the Financial Crimes AI Agent. FIS has been explicit that compliance is where the architecture gets proven, not where it stops. The same agent framework is designed to extend into the parts of banking that touch revenue directly: how credit is underwritten, how new customers are acquired, how deposits are retained and how fraud is caught. For banks paying attention, the question is not whether to engage with agentic AI. It is whether they are building the infrastructure to participate when the technology reaches the credit and origination stack.

Why AML Was the Right Place to Begin

Anti-money laundering compliance shares several characteristics with the categories of work AI agents handle best. The job is structured, rule-bound and generates an enormous volume of low-complexity decisions that consume analyst time disproportionate to their investigative complexity. A human team spending three days on a single case is not doing three days of creative judgment work. Much of it is retrieval, pattern-matching and documentation assembly. That is exactly the category of work an agent can absorb.

The governance demands are also high, which matters as much as the efficiency case. Any bank deploying an AI system in a compliance context needs full auditability, data sovereignty and a clear record of how each decision was reached. FIS built those requirements into the architecture from the start. The agent operates within FIS’s own infrastructure, no data is routed externally, and every action it takes produces a complete audit trail. That foundation does not need to be rebuilt when the program expands to credit decisioning. It carries over.

The Architecture of an Agent-First Bank

What FIS is constructing is not a point solution. It is a governed environment in which AI agents can operate across the core banking stack with full accountability. The roadmap toward credit decisioning and customer onboarding means that agents will eventually need access to the same data and decisioning frameworks that currently sit behind human underwriters, relationship managers and compliance officers. Building that environment from financial crime is sensible: the data access requirements are significant, the regulatory stakes are real and the institutional tolerance for a well-documented failure is higher than it would be in lending.

The practical implication for banks is that agent-readiness is an infrastructure question before it is a product question. A bank cannot deploy an agent that makes loan recommendations if its core systems cannot surface the right data in a machine-readable format, respond to agent queries in real time and log every decision in a form that satisfies the regulator. The FIS model addresses this at the infrastructure layer. Banks that do not have a comparable stack, whether built, licensed or partnered, will not be in a position to compete when the agent-facing lending market develops.

The Distribution Shift That Changes the Competitive Map

Yaacov Martin, CEO of Jifiti, positions agentic AI not primarily an efficiency technology for banks, but as a distribution technology – and the implications of that distinction are substantial. Speaking at FinTech Connect in 2025, Martin described the development as “embedded lending on steroids,” a reference to the way AI agents are collapsing the distance between the moment a borrower recognizes a need and the moment a loan product is placed in front of them.

Martin’s parallel is to the early internet. “If you weren’t indexed, you were nowhere.” A business that could not be found by a search engine did not compete for the customers who used one. The same dynamic now applies to AI agent ecosystems. A lending product that cannot be discovered, evaluated and executed by an AI agent will not compete for the customers whose financial lives are increasingly mediated by those agents. AI agents already influenced $262 billion in U.S. consumer sales during the 2025 holiday season alone. The trajectory toward agent-mediated financial decisions is not theoretical.

Martin’s estimate that full-scale adoption of agentic AI in financial services is approximately four years out means the preparation window is now. Banks that wait until the market has formed will find the distribution infrastructure already controlled by the institutions that moved earlier.

What Banks Need to Build Before the Credit Agents Arrive

The Financial Crimes AI Agent represents phase one of a broader transformation. The governed environment FIS is building for AML will serve as the technical and regulatory foundation for whatever comes next. When credit decisioning agents arrive, the banks positioned to benefit will be those that have already solved three problems: loan product data that is machine-readable and discoverable by agents, real-time API connectivity that allows agents to query and receive decisions without friction and a governance framework that holds up when an AI system participates in a credit decision.

The global AI agents in financial services market is projected to reach $4.28 billion by 2032. Generative AI in lending is expected to become an $8.09 billion market by 2029. Those figures are the demand side of an equation that the FIS-Anthropic partnership is beginning to address on the supply side. Banks that treat this as a technology watch item rather than a strategic infrastructure investment are underestimating the speed at which distribution in lending will change.

What Does The FIS-Anthropic Deal Mean for Banks?

FIS processes transactions for banks representing nearly 12% of global economic output. When a provider at that scale embeds Anthropic’s engineering team within its own infrastructure to build agent-first banking products, it is not experimenting. It is signaling a direction. The institutions that benefit most from that direction will be the ones whose lending operations are already built to operate within an agent ecosystem, not the ones working to retrofit their stack once the market has moved.