FIS and Anthropic Extend Their Partnership With a Claude-Based Financial Crimes Agent
Source: Businesswire — 2026-07-16
Summary
FIS and Anthropic extended their banking partnership around two initiatives: a Claude-based Financial Crimes AI Agent that handles anti-money-laundering (AML) alert triage and case investigation, and a separate security effort called Project Glasswing that uses Anthropic's Mythos 5 to help harden critical banking infrastructure. Early adopters include BMO and Amalgamated Bank, with broader availability planned for the second half of 2026.
Key Takeaways
- The Financial Crimes AI Agent handles the AML workflow end-to-end: triaging suspicious-activity alerts, investigating flagged cases, and — implicitly — assembling the resulting documentation, all built on Claude.
- Project Glasswing is a distinct security initiative under the same partnership, using Anthropic's Mythos 5 model specifically to help harden critical banking infrastructure rather than customer-facing banking workflows.
- BMO and Amalgamated Bank are named early adopters, with the offering expected to reach broader availability in the second half of 2026 — giving this a concrete adoption timeline rather than being a vague roadmap announcement.
Reel Script
Hook (~18s): Banks generate thousands of anti-money-laundering alerts a day, and most of them are false alarms that still eat a human investigator's time. FIS and Anthropic just extended a partnership to hand that triage to a Claude-based agent instead.
Core Concept (~80s): Anti-money-laundering compliance works like this: transaction-monitoring systems throw off alerts whenever something looks statistically unusual — a large transfer, an odd pattern, a mismatched profile — and a human analyst has to look at each one and decide whether it's actually suspicious or just noise. The problem is volume: the overwhelming majority of alerts turn out to be nothing, but someone still has to read each one, pull the relevant account history, and make that call, which is exactly the kind of high-volume, pattern-matching-plus-judgment work that's expensive to scale with people alone. The Financial Crimes AI Agent is built to sit in that pipeline directly — not just flagging alerts, but actually investigating them: pulling context, assessing whether a case looks genuinely suspicious, and helping build out the record if it does.
Hands-On (~110s): Picture the workflow as a three-stage pipeline you could sketch on a whiteboard. Stage one: an alert comes in from the bank's existing transaction-monitoring system, same as always. Stage two — this is the new part — instead of routing straight to a human queue, the Claude-based agent triages it first: pulling relevant account and transaction history, checking it against known typologies, and deciding whether the alert clears as routine or needs deeper investigation. Stage three: for cases that need it, the agent assists with the actual investigation and case documentation, handing a human analyst a much more complete starting point instead of a raw alert. Running alongside this on the security side is Project Glasswing, a separate track using a different Anthropic model, Mythos 5, aimed not at customer-facing workflows but at hardening the bank's own critical infrastructure — worth noting as a reminder that "AI in banking" partnerships increasingly split into a customer-workflow track and an internal-security track running in parallel.
Takeaway (~25s): This is a concrete, named-client pilot (BMO, Amalgamated Bank) with a real timeline (H2 2026), not a vague roadmap slide — which makes it a genuine early signal for where agentic AI in regulated compliance work is actually headed. If you work anywhere near AML or fincrime tooling, this partnership is worth tracking as a preview of what "good enough to deploy" looks like in a heavily regulated workflow.