FIS starts selling its AI agent for money laundering investigations
IA en un minuto newsroom · Editor: Jon Elgezabal
In 30 seconds
Banks across the United States, FIS clients or not, are now able to buy the anti-money laundering AI agent the company built with Anthropic on Claude models. Amalgamated Bank estimates it would cut the first review of alerts by more than 90% and the writing of suspicious activity reports by more than 80%. Every case is closed by an analyst, never by the machine.
FIS, the Jacksonville, Florida, company that supplies banks with their core, payments and data systems, announced on Thursday, October 15, that its AI agent for money laundering investigations is now on sale. The FIS AML Investigations Agent can be bought by any bank based in the United States, whether or not it is an FIS client. The company had unveiled it with Anthropic in May, when, according to PYMNTS, it was called the Financial Crimes AI Agent.
FIS developed it and owns it. An Anthropic team joined its product and regulatory compliance teams, and the agent runs on Claude models inside the FIS private cloud. Amalgamated Bank, one of the banks that helped design it, is already rolling it out. The joint work also yielded the evaluation and engineering methods FIS plans to use for its next agents: this is the first of those it has lined up for banking.
Stan Szwalbenest, who leads fraud and financial crimes at FIS, explains in the release that investigations stall because the evidence sits across several systems and every alert gets the same treatment. That is why, before gathering the evidence, the agent weighs each alert against the type of customer, their history and what can be expected of them. FIS gives two examples: a seasonal rise in cash deposits is not read the same way as a run of deposits kept just below the threshold for reporting to federal authorities, nor is a routine international transfer read like an unexplained payment to a high-risk country. The agent tells the analyst why it separates one case from another.
The time savings are a projection by Amalgamated, which compared its current processes with what the agent is expected to do: more than 90% less on the first review of alerts (known as level 1), about 98% less on the in-depth study of escalated cases (level 2) and more than 80% less on writing suspicious activity reports.
What the agent produces are recommendations: no case is closed unless an analyst expressly confirms it. Each statement points to the record, document or cheque image it comes from, and the recommendation, its reasoning and the final decision are all stored for internal audits and supervisory inspections. Each bank's risk rules, thresholds and procedures are kept apart from everyone else's.
The agent does not replace the monitoring platforms or case management tools each bank already uses, but works alongside them. It connects natively with FIS Horizon and FIS Systematics, two of the company's core banking platforms, and the next one on its roadmap is FIS IBS.
To show the size of the problem, FIS cites a 2023 LexisNexis estimate: U.S. financial institutions spend between $35 billion and $40 billion a year on their anti-money laundering operations. According to other studies quoted in the release, analysts spend between 40% and 60% of each investigation pulling data together (AnChain.AI, 2025) and between 90% and 95% of the industry's alerts turn out to be false positives (Datos Insights, 2023). The release gives no price and does not say whether the agent will reach banks outside the United States.
Why it matters · analysis and opinion
Having it sold by the provider that already runs the core of the accounts changes the conversation with banks: the agent does not need to ask for access to the data, because it works where the data already lives. It also shows how Anthropic is entering regulated banking, alongside a veteran partner that owns the product. The time savings are one bank's projection, not results measured in production, and should be read that way. For compliance departments, the real test will be whether supervisors accept reports prepared by an AI and signed by the analyst; logging every step points squarely at that requirement. Outside the United States, for now, it cannot be bought.
Official source: FIS · Written with the help of AI: how we make the news
