At Semafor's Banking on the Future Forum, Revolut U.S. CEO Cetin Duransoy said the fintech's AI-driven transaction-monitoring systems perform "statistically significantly better than human reviews of the transactions," according to CUToday.info reporting on American Banker. PYMNTS reports Revolut's compliance stack operates across 39 countries, with agentic AI supporting both know-your-customer onboarding and ongoing transaction monitoring and humans concentrating on higher-risk cases. CUToday.info also reports Duransoy warned that fraudsters are deploying agentic AI to create convincing fake personas, and he voiced support for the proposed U.S. Scam Act to increase platform advertiser verification. These comments describe Revolut's public statements at the forum; reporting does not publish the underlying model names, performance metrics, or independent audit results.
What happened
Cetin Duransoy, Revolut's U.S. CEO, told attendees at Semafor's Banking on the Future Forum that the company's AI-driven transaction monitoring is "statistically significantly better than human reviews of the transactions," according to CUToday.info's coverage of American Banker. PYMNTS reports that Revolut's compliance stack runs across 39 countries and that the firm uses agentic AI for both know-your-customer onboarding and ongoing transaction monitoring, while human investigators handle higher-risk cases, per PYMNTS reporting.
Technical details
PYMNTS describes Revolut's architecture as a layered compliance stack where high-volume, lower-complexity screening is handled by AI and humans are routed the cases requiring judgment. PYMNTS additionally notes industry pressure from real-time payments systems and changing limits on instant-transfer rails that demand faster AML decisioning. The public reporting does not specify model names, training data, evaluation methodology, or precise performance metrics beyond the quote attributed to Duransoy.
Industry context
Editorial analysis: Companies operating large-scale transaction-monitoring programs have increasingly explored machine learning to reduce false positives and meet real-time decision windows. Reporting cited by PYMNTS places industry AML compliance costs above $274 billion annually and highlights legacy rule-based systems' high false-positive rates, which industry observers say motivated more automation experiments.
Policy and risk signals from the forum
CUToday.info reports Duransoy warned about an 'arms race' as fraudsters deploy agentic AI to create fake personas and run automated scams. Per CUToday.info, he also expressed support for the proposed U.S. Scam Act to require greater advertiser verification on social platforms. The coverage does not include an independent assessment of how Revolut's systems mitigate risks such as model bias, adversarial manipulation, or operational failure modes.
What to watch
For practitioners: look for independently verifiable performance data, precision, recall, false-positive reduction, and time-to-decision, and for audit or regulatory filings that document model governance. Also follow legislative developments around the proposed Scam Act and any industry guidance on AI use in AML, since those could change compliance requirements and third-party risk expectations.
Limitations of the reporting
What is reported here is based on public remarks and trade reporting. Neither PYMNTS nor CUToday.info published model-level metrics, technical papers, or independent audits tied to the claim that the AI is "statistically significantly better" than human reviewers.
Scoring Rationale #
A notable operational claim about AI outperforming human AML reviewers matters to compliance and ops teams, but reporting lacks independent metrics or audits. The story signals industry momentum on ML for AML rather than a verified, sector-changing result.
Practice with real Payments data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card