Skip to main content
Loading navigation
Back to challenges

Stablecoin Settlement Guardrails with OpenAI Agents SDK and GPT-5 Pro

Emerging stablecoin payment rails require real-time compliance validation and automated fraud screening before final on-chain settlement. You will build an autonomous settlement verification agent that evaluates cross-border transaction risks and enforces strict sanctions policy rules, targeting sub-second decision latencies and zero compliance leaks.

CybersecurityHosted by Vera
Sign in free to start

What you are building

Build an automated stablecoin settlement guardrail system using OpenAI Agents SDK and GPT-5 Pro to audit transactions against dynamic sanctions lists.

How work is evaluated

Evaluation

Evaluates stablecoin settlement approval logic and compliance adherence under synthetic high-volume transaction stress.

Datasets

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

How submissions are scored

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

zero_false_negatives

Ensures no sanctioned transaction is approved

Binary check

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

Dimension 2

accuracy

Overall accuracy of settlement risk classification • target: 0.99 • range: 0.95-1

Binary check

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

Learning goals

  • Implement real-time transaction screening using OpenAI Agents SDK

  • Configure GPT-5 Pro tool functions to query sanctions and entity resolution databases

  • Deploy sub-second latency transaction policy verification workflows

  • Validate zero-leak compliance audit trails for stablecoin transfers

Resources

Dataset notes

Synthetic stablecoin settlement stream containing 10,000 transaction records with embedded sanctions list matches and anomaly flags.

Challenge setup

Evaluates stablecoin settlement approval logic and compliance adherence under synthetic high-volume transaction stress.

Challenge input

JSON containing sender, receiver, amount, and asset details

OpenAI Agents SDK

Core orchestration framework for agent tool calls and structured reasoning.

GPT-5 Pro

High-reasoning capability model for evaluating complex risk profiles.

Evaluated output

JSON with decision ('APPROVE'|'REJECT'), reason, and confidence score

Evaluation checks
  • Ensures no sanctioned transaction is approved
  • Overall accuracy of settlement risk classification • target: 0.99 • range: 0.95-1
Configured evaluation evidence
  • Benchmark: FinCrime-Eval-v1
  • Accuracy target: 0.99
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Shows the first task, up to three tools, and up to three evidence items.

View technical recipe

Configured tools

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

Evaluation contract

  • zero_false_negatives · Weight 1
  • accuracy · Weight 1

Metrics

  • accuracy · Target 0.99

Benchmarks

  • FinCrime-Eval-v1
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing sender, receiver, amount, and asset details

    Output: JSON with decision ('APPROVE'|'REJECT'), reason, and confidence score

    Execution harness: Python

Evaluation tools

  • OpenAI Agents SDK · agent_framework
  • GPT-5 Pro · llm_engine

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Stablecoin Settlement Guardrails with OpenAI Agents SDK and GPT-5 Pro