From Fraud Score to Fraud Case: Agents Now Do the Investigation
Fraud AI is moving past "risk score 87". Agentic systems now gather the evidence, build the case and hand a recommendation to a human.
"Risk score: 87" isn't an answer. It's homework. In 2026, agents are doing the homework.
📰 THE NEWS
FIS and Anthropic have introduced a Financial Crimes AI Agent that gathers data across a bank's core systems, checks suspicious activity against known risk patterns and prioritises high-risk cases for human investigators. Bank of Montreal and Amalgamated Bank are early adopters, with wider availability planned for the second half of 2026. McKinsey research finds early agentic use cases in KYC and AML cutting manual workload by 30–50%.
The other side is upgrading too: the 2026 INTERPOL Global Financial Fraud Threat Assessment, cited by Thomson Reuters, says agentic AI has made fraud schemes more than four times as profitable as the digital methods before them.
🔄 THE PIVOT
Old model: alert → analyst spends two hours gathering evidence → decision.
New model: alert → AI investigates and assembles the context → recommendation → human approves.
The analyst becomes the judge, not the researcher.
✅ ACTIONABLE TIP for AP, treasury and controllers
1. List your top three alert types by volume (duplicate payments, vendor bank-detail changes, unusual journals).
2. For each, write down the checklist your best investigator follows today. That's your agent's playbook.
3. Let AI assemble the evidence pack: vendor history, approval trail, bank-change log, similar past cases.
4. Keep the action human: block, release or escalate stays with a named approver.
5. Log every AI recommendation next to the final human decision. Your auditors will ask.
💬 How long does your team take to investigate one payment alert today: minutes, hours or days?
👉 Join Free at www.FinanceClub.Ai
#FraudDetection #AgenticAI #FinancialCrime #AML #InternalControls #AccountsPayable #AIinFinance #FinanceClubAi
Sources: FIS and Anthropic announcement (via IT Digest); Backbase (May 2026) citing McKinsey; Thomson Reuters white paper on agentic AI for investigations.