FIS AML Investigations Agent Now Available to US Banks

FIS announced on 15 October 2026 that its AML Investigations Agent — built by FIS with support from Anthropic and first jointly announced in May 2026 — is now commercially available to any U.S.-based bank, including banks that are not FIS clients today.

The agent runs in FIS’ private cloud and uses Anthropic’s Claude models. FIS says it is the first in the company’s roadmap of AI agents for banking, and that it is now being deployed at Amalgamated Bank, a design partner.

Projected time savings — Amalgamated’s analysis, not measured results

FIS cites projections based on Amalgamated’s analysis of its current workflows against the agent’s expected abilities. Those projections are not published measured before-and-after results:

Level 1 alert review time: cut by more than 90 percent
Level 2 case investigation time: cut by approximately 98 percent
SAR narrative preparation time: cut by more than 80 percent

FIS defines Level 1 as initial alert triage and Level 2 as deeper case review after escalation. Treat every percentage as a bank/company projection.

Controls: recommendations only, analyst confirmation required

Built on FIS Trust Assured, the company says every output is a recommendation, no case can be closed without explicit analyst confirmation, and each statement links to its source record. Each institution’s risk logic, procedures and thresholds remain isolated through the FIS Control Plane. Recommendations, reasoning and the analyst’s final decision are logged in an audit trail for internal audit or regulatory examination.

Sean Searby, Amalgamated’s chief information and operations officer, said the agent can reduce time spent gathering information and preparing documentation, while “the investigator remains accountable for the final determination.”

Stan Szwalbenest, FIS head of enterprise fraud and financial crimes, said: “Investigations don’t slow down because analysts lack judgment. They slow down because evidence is scattered across systems and every alert gets treated the same.”

How FIS describes the workflow

Rather than treating every alert the same, FIS says the agent maps the customer’s type, historical activity and expected behaviour against the alert before assembling evidence — so a seasonal cash-deposit spike is framed differently from structured deposits under reporting thresholds, and a routine international transfer differently from an unexplained high-risk wire. The agent is designed to augment, not replace, existing monitoring and case-management systems, and natively integrates with FIS Horizon and FIS Systematics, with FIS IBS planned next.

FIS developed and owns the agent; an Anthropic team embedded with FIS product and compliance teams. The collaboration also produced evaluation frameworks FIS says it will reuse for more agents on the same governed platform.

Industry figures — credited as FIS credits them

FIS cites three third-party industry figures to frame the problem (credit each source as FIS does):

U.S. AML operations cost an estimated $35–$40 billion a year (LexisNexis, 2023)
Analysts spend 40–60 percent of investigation time assembling data before analysis (AnChain.AI, 2025)
Industry false-positive rates run 90–95 percent (Datos Insights, 2023)

Those are industry estimates FIS quotes, not FIS’s own measurements of Amalgamated’s book.

Why it matters

AML investigation is a classic “evidence assembly before judgment” bottleneck. An agent that sits on core banking, payments and data infrastructure — with human confirmation still mandatory — is the enterprise pattern regulators are more likely to tolerate than a black-box closer. Commercial availability beyond existing FIS clients also widens who can buy it.

For adjacent Anthropic enterprise partnerships, see HCLTech’s global premier partnership with Anthropic. For regulatory pressure on financial-crime controls in crypto specifically, see TSN’s UK FCA crypto licensing regime coverage — a different regime, same compliance-capacity theme.

What’s still unknown

Measured results once Amalgamated (or others) publish live metrics
Pricing and whether non-FIS cores need custom connectors beyond Horizon/Systematics
How Claude model versions are pinned and updated inside FIS’ private cloud

What this means

FIS AML Investigations Agent commercially available to any U.S. bank (FIS).
Built with Anthropic, runs on Claude in FIS’ private cloud; deploying at Amalgamated.
Time cuts (>90% / ~98% / >80%) are Amalgamated projections, not results.
Industry cost/false-positive stats: credit LexisNexis / AnChain.AI / Datos Insights as FIS does.