New York, May 5: Somewhere in a compliance department right now, an analyst is staring at a screen full of flagged transactions. Most of them are nothing. A few might matter. The problem is telling the difference, and doing it fast enough, and doing it across thousands of accounts, every single day, without burning through people who will eventually quit because the work is relentless and the tools are, frankly, not good enough.
- The AI Agent Financial Crime Problem That Banks Have Been Too Polite to Talk About
- Why This Pairing Makes Sense Beyond the Press Release
- BMO and Amalgamated Bank Are First. The Rest of the Industry Is Watching.
- The Roadmap Behind the First Agent Is Where This Gets Interesting
- What the Analysts Who Sit in Those Compliance Rooms Actually Need
That is the problem Anthropic and FIS announced they are building toward on Monday, May 4. The two companies went public with a Financial Crimes AI Agent, a purpose-built system designed to take over the investigative grunt work that has quietly consumed banking compliance departments for decades. The Wall Street Journal first reported the partnership, and FIS followed with a full press release out of Jacksonville confirming the scope. Shares of FIS jumped roughly 7 percent in after-hours trading. The market, at least, thought it meant something.
The AI Agent Financial Crime Problem That Banks Have Been Too Polite to Talk About
Ask any compliance officer at a mid-sized bank what their biggest operational headache is, and most of them will not say regulation. They will say the gap between what their current systems can flag and what their teams can actually investigate. Those two numbers have never matched, and the distance between them keeps growing.
According to figures cited in the FIS press release, US financial institutions collectively spend somewhere between $35 and $40 billion every year on anti-money laundering operations. The United Nations puts the volume of illicit funds moving through the global financial system at roughly $2 trillion annually. The detection rate, per the Financial Action Task Force, sits at less than one percent of that. Spend a lot, catch very little. That is the industry’s quiet embarrassment, and almost nobody outside the compliance world fully grasps how structurally broken the current model is.
Part of the problem is purely mechanical. Financial crime data, as FIS described it in Monday’s announcement, sits locked in disconnected systems. A transaction flag might live in one platform, the customer’s account history in another, a prior case note somewhere else entirely. An investigator trying to build a picture of suspicious activity has to manually pull from multiple sources, cross-reference them, and write up what they find in a Suspicious Activity Report narrative that will go to FinCEN and potentially end up in front of a federal prosecutor. That process takes hours, sometimes days, per case.
The agent FIS and Anthropic have built is designed to do that assembly work automatically. It pulls evidence across a bank’s core systems, evaluates the activity against known typologies for money laundering and financial crime, and surfaces the cases most worth a human investigator’s attention. The investigators still make the calls. The agent does the digging.
Why This Pairing Makes Sense Beyond the Press Release
Anthropic is an interesting choice of partner here, and not just because Claude is a capable model. The company, founded in 2021 and now valued north of $61 billion according to prior reporting by The Wall Street Journal, has spent years building its models with an emphasis on what it calls safety and interpretability. In banking compliance, that is not a branding preference. It is a legal requirement.
When a Suspicious Activity Report ends up supporting a prosecution, every piece of reasoning behind it needs to hold up to scrutiny. A model that produces confident-sounding outputs without a clear link to the underlying evidence creates serious exposure for the institution that filed the report. Anthropic’s Head of Financial Services Jonathan Pelosi addressed this directly in Monday’s release, noting that the team was embedded inside FIS specifically to ensure that every conclusion the agent reaches links back to its source data, and every decision stays with the investigator.
That architecture matters more than it sounds. FIS is not building a black box. The agent-first governed environment being constructed keeps client data within FIS-controlled infrastructure, with every agent decision designed to be traceable and auditable. For institutions operating under consent orders or active regulatory monitoring, that is the only configuration that even gets through procurement.
FIS itself brings something Anthropic cannot supply alone: access. FIS processes transactions for financial institutions representing nearly 12 percent of the global economy, serving as the system of record for transactions, payments, deposits, credit, and customer activity across thousands of banks. That positioning means the agent does not need new integrations to access the data investigators need. The infrastructure is already there. For institutions not running FIS core systems, open integration standards provide the connection path.
Turns out, when you want to build something that actually works inside real banking operations, picking a partner who already lives inside those operations is the more pragmatic move.
BMO and Amalgamated Bank Are First. The Rest of the Industry Is Watching.
Bank of Montreal and Amalgamated Bank will be among the first institutions to use the financial crimes agent, with broad availability expected for FIS clients in the second half of 2026. Neither bank is the largest institution in the room, which is actually by design. Proving the model works at scale requires running it in real production environments, not just controlled pilots. BMO brings international correspondent banking complexity. Amalgamated, a smaller institution with a socially conscious depositor base, brings a different risk profile entirely. Together they offer a reasonable stress test of whether the agent performs consistently across different institutional contexts.
What happens in those deployments will be closely tracked across the industry. The compliance technology market is competitive and not short on promises. NICE Actimize, Quantexa, Hawk AI, and others have been building in adjacent spaces, and the major core banking vendors have all been talking about AI integration for years. The difference here is that FIS is not talking about integration. It is talking about an agent already in development with named clients and a specific H2 2026 go-live timeline.
FIS CEO Stephanie Ferris was direct about the direction of travel, stating that every bank in the world wants AI that acts, not just assists, and that FIS built the architecture to orchestrate this intelligence with Claude as the reasoning engine inside. That is a confident posture from a company whose stock has dropped roughly 40 percent over the past twelve months, partly on investor concerns that AI could erode demand for legacy software providers. The FIS-Anthropic partnership is, in part, a counter-argument to that thesis.
The Roadmap Behind the First Agent Is Where This Gets Interesting
Financial crime compliance is the opening act. The roadmap FIS outlined spans credit decisioning, deposit retention, customer onboarding, and fraud prevention. Those are not adjacent use cases. They are core banking functions, and deploying AI agents across all of them represents a fundamental shift in how bank operations get executed.
The knowledge transfer element built into the Anthropic partnership structure is worth noting here. Anthropic’s Applied AI team and forward-deployed engineers are embedded with FIS to co-design the financial crimes agent, with an explicit goal of transferring knowledge so FIS can build and scale additional agents independently over time. That is not a typical vendor relationship. It looks more like Anthropic training FIS to fish rather than selling them fish indefinitely. Whether that structure holds as the relationship matures is a legitimate question, but the intent signals that FIS is serious about owning its own AI capability, not just licensing someone else’s.
For banks considering their own AI compliance investments, that structure also raises a question worth sitting with: at what point does building directly on a foundation model provider, inside a governed infrastructure layer you already trust, become more defensible than running a procurement process for a purpose-built compliance software vendor? The answer is not obvious yet. But the question is arriving faster than most compliance teams expected.
What the Analysts Who Sit in Those Compliance Rooms Actually Need
There is a version of this story that stays at the level of press releases and stock movements and misses what actually matters. What actually matters is the person doing the investigation.
Compliance analysts at most financial institutions are not underskilled. They are under-tooled and overwhelmed. False positive rates on traditional transaction monitoring systems run so high that the alert review process becomes, functionally, a filter for looking busy rather than a genuine detection mechanism. When the signal-to-noise ratio is that bad, experienced investigators start discounting alerts instinctively, and that is exactly when real cases slip through.
An agent that genuinely reduces false positives, assembles the relevant evidence before the investigator even opens the case, and drafts a coherent narrative that a human can verify and sign off on: that changes the job. Not eliminates it. Changes it. The investigator spends less time pulling data and more time actually thinking about whether the pattern in front of them represents something real. That is a meaningful upgrade.
FIS stated the agent will be evaluated on reducing cost per case, cutting manual work, and decreasing case review time, with investigators retaining control of final decisions. Those are the right metrics. They are also the metrics that will determine whether this stays a good press announcement or becomes a genuine shift in how the industry operates.
The TD Bank settlement in late 2024, the largest bank money laundering penalty in US history at the time, made clear to every institution in the country that the regulatory appetite for compliance technology gaps had run out. Banks can no longer treat AML infrastructure as a compliance checkbox and expect that to be enough. The expectation now, from regulators at the OCC, FinCEN, and the Federal Reserve, is that institutions adopt systems capable of actually working at the speed and volume of modern transaction flows.
Anthropic and FIS are building directly at that gap. The first deployments will show whether what they have built is ready for what the job actually demands.
The banks waiting to see those results should not wait too long.
Connect With Us On Social Media [ Facebook | Instagram | Twitter | LinkedIn ] To Get Real-Time Updates On The Market. Entrepreneurs’ Diaries Is Now Available On Telegram. Join Our Telegram Channel To Get Instant Updates.
Isabella is a global business journalist and former McKinsey analyst from Brazil. She brings sharp insights on economic shifts, policies, and founder journeys from around the world.
Luca is a tech ethicist from Italy exploring disruptive innovation through a human lens—from AI to biotechnologies to decentralization.
