Skip to main content
Back to challenges
Challenge

Multi-Agent Cross-Border Payment Sanctions Triage Network Using AutoGen and GPT-5 Pro

Global financial institutions processing ISO 20022 payment messages through SWIFT network nodes require multi-perspective verification agents to resolve ambiguous sanctions matches without stalling legitimate settlement corridors. Using AutoGen and GPT-5 Pro, developers can build a collaborative multi-agent triage swarm where specialist agents debate entity resolution, sanctions jurisdiction, and false-positive likelihood prior to final release.

CybersecurityHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Build a multi-agent conversational compliance network using AutoGen and GPT-5 Pro to resolve high-risk cross-border SWIFT payment sanctions alerts.

Delivery guide

How work is evaluated

Evaluation

Evaluates multi-agent conversation convergence, consensus accuracy, and final transaction verdict generation on complex ISO 20022 payment cases.

Datasets

Challenge datasets

Public datasets, plus the private datasets you have access to.

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

multi_agent_consensus_reached

All participating agents in group chat must agree on final verdict.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

sanctions_triage_accuracy

Accuracy of payment release vs freeze decisions compared to expert gold standard. • target: 0.96 • range: 0.88-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

What you should walk away with

  • Orchestrate multi-agent group chats using AutoGen with distinct specialist roles

  • Leverage GPT-5 Pro for high-reasoning legal analysis and complex entity disambiguation

  • Implement structured ISO 20022 message parsing and field-level risk attribution

  • Design consensus-driven resolution protocols that minimize false positive holds

Resources and assets

Reference links and supporting material

Dataset notes

Dataset of 400 ISO 20022 message payloads involving complex cross-border payment paths across UAE, UK, Singapore, and EU rails with deliberate sanctions ambiguity.

Challenge setup

Evaluates multi-agent conversation convergence, consensus accuracy, and final transaction verdict generation on complex ISO 20022 payment cases.

Challenge input

JSON representing an ISO 20022 pacs.008 transaction with nested party details.

AutoGen

Enables multi-agent debates and consensus formation for complex compliance cases.

GPT-5 Pro

Delivers advanced multi-step reasoning for cross-border legal and sanctions analysis.

Evaluated output

JSON with consensus_verdict (RELEASE|FREEZE|ESCALATE), debate_rounds, and unanimous_reasoning.

Evaluation checks
  • All participating agents in group chat must agree on final verdict.
  • Accuracy of payment release vs freeze decisions compared to expert gold standard. • target: 0.96 • range: 0...
Configured evaluation evidence
  • Benchmark: Multi-Agent Compliance Swarm Bench
  • Sanctions Triage Accuracy target: 0.96
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.

View technical recipe

Configured tools

Action Space
  • AutoGen · Required
  • GPT-5 Pro · Optional
Policy Serving
  • GPT-5 · Optional

Evaluation contract

  • multi_agent_consensus_reached · Weight 1
  • sanctions_triage_accuracy · Weight 1

Metrics

  • sanctions_triage_accuracy · Target 0.96

Benchmarks

  • Multi-Agent Compliance Swarm Bench
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON representing an ISO 20022 pacs.008 transaction with nested party details.

    Output: JSON with consensus_verdict (RELEASE|FREEZE|ESCALATE), debate_rounds, and unanimous_reasoning.

    Execution harness: Python

Evaluation tools

  • AutoGen · Multi-Agent Conversational Framework
  • GPT-5 Pro · Reasoning Model Engine

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Multi-Agent Cross-Border Payment Sanctions Triage Network Using AutoGen and GPT-5 Pro