--- title: "FIS and Anthropic Cut AML Investigations by 90%. The AI Model Wasn't the Hard Part." date: 2026-08-08T09:00:00Z author: Jason Guest url: https://truehorizon.ai/news/fis-anthropic-financial-crimes-agent-compliance-moat description: "FIS and Anthropic cut manual AML work by 90% with a new AI agent. Here's why the model itself wasn't the hard part." --- # FIS and Anthropic Cut AML Investigations by 90%. The AI Model Wasn't the Hard Part. > FIS and Anthropic cut manual AML work by 90% with a new AI agent. Here's why the model itself wasn't the hard part. FIS and Anthropic just gave enterprise leaders a real data point on what agentic AI looks like inside a regulated industry. The two companies built a Financial Crimes AI Agent that automates anti-money-laundering investigation work, and in early testing it cut manual process work by as much as 90%. That number is the headline. The more useful story is buried a few paragraphs into the [Forbes piece](https://www.forbes.com/sites/nicolecasperson/2026/05/06/fis-and-anthropic-signal-a-new-era-of-ai-infrastructure-in-banking/) that broke it: FIS's own CEO says the AI model was never the hard part. ## The 90% number, and what it actually covers The Financial Crimes AI Agent automatically assembles evidence across a bank's core systems, checks it against known money-laundering patterns, and surfaces the highest-risk cases for a human investigator to review. In early testing, FIS says that cut manual investigation work by as much as 90%. The scale of the problem it targets is enormous: the UN estimates roughly $2 trillion moves through the global financial system illicitly every year, and US institutions alone spend $35 to $40 billion annually trying to catch it. Anti-money-laundering compliance isn't a niche cost center. It's one of the most expensive ongoing processes in banking, which is exactly why a 90% reduction in manual work is worth paying attention to. ## Why the model wasn't the moat Here's the part most coverage buried. FIS CEO Stephanie Ferris was blunt about where the real value sits: "Our moat is the 50 years of deep regulatory and compliance experience. The hard part is, is it compliant? Is it correct every time, with every regulation? You're not going to innovate around us." Anthropic supplied the model. FIS supplied the fifty years of knowing exactly what being correct every time, with every regulation actually requires in banking. That distinction matters far beyond this one partnership. Even Anthropic didn't claim otherwise. The company's Head of Financial Services described the requirement as a model that could reason through complex investigations accurately, not a model that already understood banking compliance out of the box. That knowledge came from FIS's side of the partnership. When the company that built the model says the compliance layer still has to come from somewhere else, that's worth taking at face value. ## The partnership already expanded This wasn't a one-off pilot. By July, [Crowdfund Insider reported](https://www.crowdfundinsider.com/2026/07/292587-fis-and-anthropic-extend-partnership-to-deploy-advanced-ai-agents-in-banking/) that the FIS-Anthropic partnership had grown from the original financial crimes agent into a broader roadmap covering credit decisioning, customer onboarding, fraud prevention, and deposit retention, all inside the same governed AI platform. Ten weeks is a fast timeline for that kind of expansion. Shaky pilots don't usually earn a bigger scope that quickly. Deployments that hold up under regulatory scrutiny do. ## What this means if you're evaluating AI vendors The lesson here isn't specific to banking. Any regulated or high-stakes function considering agentic AI should take the same read from this partnership: the model is necessary but not sufficient. Three things worth doing before you sign anything: - Ask what makes the deployment correct every time, with every regulation, not just which model is running underneath it. - Look for evidence the deployment is scaling into new use cases, not just one strong pilot metric. - Bring in people who have actually built compliance-grade AI systems before. That expertise took FIS 50 years to build, and it's the part most AI rollouts skip. ## Where TrueHorizon fits This is exactly the gap we work in. We've built agentic AI deployments where being correct every time matters as much as being capable, which is a different discipline than shipping a capable model and hoping the compliance question sorts itself out later. That's not a checklist we're learning on your project. It's the expertise we bring to it. If the FIS and Anthropic partnership proves anything, it's that the model is now the easy part to buy. The hard part, the part that actually determines whether an AI deployment survives contact with a regulator, is exactly what we specialize in. If you're evaluating agentic AI for a function where getting it wrong has real consequences, [take our AI readiness assessment](https://truehorizon.ai/assessment) before you pick a vendor. Source: https://truehorizon.ai/news/fis-anthropic-financial-crimes-agent-compliance-moat