Crypto Banking Off-Ramp AML Verification with Claude Agents SDK
Crypto-to-fiat banking off-ramps require rigorous multi-jurisdictional AML source-of-funds verification to satisfy banking partner compliance mandates. Build an off-ramp AML verification pipeline using Claude Agents SDK to trace wallet lineage, analyze fiat destination account legitimacy, and flag suspicious transactions with 96% accuracy.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an intelligent crypto off-ramp verification pipeline utilizing Anthropic Claude Agents SDK and tool use features.
How work is evaluated
Evaluates accuracy of wallet risk scoring and validity of generated SAR evidence packets.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
evidence_presence_check
Ensures frozen transactions contain non-empty evidence arrays.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
aml_detection_accuracy
Accuracy in correctly flagging suspicious vs legitimate off-ramps. • target: 0.96 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Construct autonomous verification workflows using Anthropic Claude Agents SDK
Implement multi-hop blockchain wallet taint analysis tools for off-ramp requests
Validate fiat destination bank details against global IBAN and SWIFT registers
Generate human-readable investigator audit trails detailing risk scoring rationale
Reference links and supporting material
Synthetic dataset of 400 crypto-to-fiat transfer requests with simulated wallet transaction graphs and bank account verifications.
Challenge setup
Evaluates accuracy of wallet risk scoring and validity of generated SAR evidence packets.
Challenge input
JSON transaction request with origin wallet address, asset, fiat payout bank, and amount
Claude Agents SDK
Anthropic's SDK supporting tool use and deep reasoning capability.
Evaluated output
JSON verification object with risk_tier, decision, and detailed_evidence array
- Ensures frozen transactions contain non-empty evidence arrays.
- Accuracy in correctly flagging suspicious vs legitimate off-ramps. • target: 0.96 • range: 0-1
- Benchmark: CryptoAML-Bench-v1
- Aml Detection Accuracy target: 0.96
- 1 public reference case
- 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
- Claude Code · Required
- Anthropic · Optional
- Claude Code · Required
- Claude 4.1 Opus · Optional
Evaluation contract
- evidence_presence_check · Weight 1
- aml_detection_accuracy · Weight 1
Metrics
- aml_detection_accuracy · Target 0.96
Benchmarks
- CryptoAML-Bench-v1
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON transaction request with origin wallet address, asset, fiat payout bank, and amount
Output: JSON verification object with risk_tier, decision, and detailed_evidence array
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.