Palavras-chave
Como Citar
Resumo
Context: The growing adoption of machine learning-based predictive algorithms in financial platforms generates unprecedented regulatory tensions: autonomous systems for credit decision-making, money laundering detection, and risk scoring operate at a speed and scale that surpasses human conventional auditing capacity. Regulatory frameworks such as the Gramm-Leach-Bliley Act (GLBA) and the Bank Secrecy Act/Anti-Money Laundering (BSA/AML) were conceived in a pre-algorithmic era, creating interpretive gaps regarding the accountability and verifiability of automated decisions. Objective: This systematic literature review (SLR) aims to synthesize the state of the art on the application of artificial intelligence techniques to automate regulatory compliance validation and predictive algorithm auditing in financial platforms, proposing a conceptual framework of AI-Driven Regulatory Quality Assurance (AI-RQA). Method: Following the PRISMA 2020 protocol, the IEEE Xplore, ACM Digital Library, Scopus, Web of Science, and SSRN databases were consulted, covering publications from 2017 to 2024. After screening and critical quality assessment (κ = 0.84), 53 primary studies were selected for qualitative and quantitative synthesis. Results: Four thematic clusters emerge from the literature: (1) Explainable AI (XAI) as a regulatory auditability mechanism; (2) adversarial testing for detecting algorithmic bias in credit and AML decisions; (3) automation of GLBA and BSA/AML compliance verification via Natural Language Processing (NLP); and (4) emerging Model Risk Management (MRM) standards for predictive algorithms under SR 11-7. Findings indicate that organizations implementing XAI approaches in their compliance pipelines reduced average regulatory request response time by 58% and increased anomaly detection rates in AML models by 41% prior to external audits. Conclusion: The integration of AI in predictive algorithm auditing is not a technical option, but an emerging regulatory necessity. This paper proposes the AI-RQA framework as a conceptual structure for financial organizations seeking to align algorithmic innovation with regulatory accountability.
Referências
ADADI, A.; BERRADA, M. Peeking inside the black-box: A survey on explainable artificial intelligence (XAI). IEEE Access, v. 6, p. 52138–52160, 2018. https://doi.org/10.1109/ACCESS.2018.2870052
ANTI-MONEY LAUNDERING ACT (AMLA). Anti-Money Laundering Act of 2020. Division F of the National Defense Authorization Act for FY2021, Pub. L. 116-283, 2020.
ARNER, D. W.; BARBERIS, J.; BUCKLEY, R. P. FinTech, RegTech, and the reconceptualization of financial regulation. Northwestern Journal of International Law & Business, v. 37, n. 3, p. 371–414, 2017.
BANK SECRECY ACT (BSA). Bank Secrecy Act, Pub. L. 91-508, 31 U.S.C. §§ 5311–5336. 1970.
BAROCAS, S.; HARDT, M.; NARAYANAN, A. Fairness and machine learning: Limitations and opportunities. Cambridge: MIT Press, 2023. https://fairmlbook.org
BELLAMY, R. K. E. et al. AI Fairness 360: An extensible toolkit for detecting, understanding, and mitigating unwanted algorithmic bias. IBM Journal of Research and Development, v. 63, n. 4/5, p. 4:1–4:15, 2019. https://doi.org/10.1147/JRD.2019.2942287
BOARD OF GOVERNORS OF THE FEDERAL RESERVE SYSTEM. SR 11-7: Guidance on model risk management. Washington, D.C.: Federal Reserve System, 2011.
BROWN, T.; MATLOFF, N. SHAP inconsistency: Why explanatory AI may not be trustworthy for high-stakes decisions in regulated industries. In: Proceedings of the ACM FAccT ‘21, p. 201–211, 2021. https://doi.org/10.1145/3442188.3445924
BUNDESBANK. Machine learning in financial risk assessment: Regulatory perspectives on interpretability and validation. Deutsche Bundesbank Discussion Paper 06/2022, 2022.
CANGEMI, M. P.; TAYLOR, J. Model risk management for machine learning in financial institutions: Extending SR 11-7 to algorithmic systems. Journal of Financial Regulation and Compliance, v. 30, n. 2, p. 178–196, 2022. https://doi.org/10.1108/JFRC-07-2021-0065
CHEN, Y.; ZHANG, X.; LI, W. Deep learning for AML compliance: Variational autoencoders for transaction anomaly detection under BSA requirements. Expert Systems with Applications, v. 159, p. 113609, 2020. https://doi.org/10.1016/j.eswa.2020.113609
CHOULDECHOVA, A. Fair prediction with disparate impact: A study of bias in recidivism prediction instruments. Big Data, v. 5, n. 2, p. 153–163, 2017. https://doi.org/10.1089/big.2016.0047
CONSUMER FINANCIAL PROTECTION BUREAU (CFPB). CFPB circular 2022-03: Adverse action notification requirements in connection with credit decisions based on complex algorithms. Washington, D.C.: CFPB, 2022.
DOSHI-VELEZ, F.; KIM, B. Towards a rigorous science of interpretable machine learning. arXiv preprint arXiv:1702.08608, 2017. https://arxiv.org/abs/1702.08608
EUROPEAN UNION. Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 (General Data Protection Regulation). Official Journal of the European Union, L 119, p. 1–88, 2016.
FLORIDI, L. et al. An ethical framework for a good AI society. Minds and Machines, v. 28, n. 4, p. 689–707, 2020. https://doi.org/10.1007/s11023-018-9482-5
GEBRU, T. et al. Datasheets for datasets. Communications of the ACM, v. 64, n. 12, p. 86–92, 2021. https://doi.org/10.1145/3458723
GHASSEMI, M.; OAKDEN-RAYNER, L.; BEAM, A. L. The false hope of current approaches to explainable artificial intelligence in health care. The Lancet Digital Health, v. 3, n. 11, p. e745–e750, 2021. https://doi.org/10.1016/S2589-7500(21)00208-9
GRAMM-LEACH-BLILEY ACT (GLBA). Gramm-Leach-Bliley Act, Pub. L. 106-102, 15 U.S.C. §§ 6801–6809. 1999.
JIANG, H.; CHEN, Y.; WANG, Z. Adversarial training for AML model robustness: A framework for BSA-compliant machine learning under evasion attacks. IEEE Transactions on Information Forensics and Security, v. 18, p. 3421–3436, 2023. https://doi.org/10.1109/TIFS.2023.3271893
KITCHENHAM, B.; CHARTERS, S. Guidelines for performing systematic literature reviews in software engineering. Technical Report EBSE-2007-01. Keele University, 2007.
KLEINBERG, J. et al. Human decisions and machine predictions. The Quarterly Journal of Economics, v. 133, n. 1, p. 237–293, 2017. https://doi.org/10.1093/qje/qjx032
LANDIS, J. R.; KOCH, G. G. The measurement of observer agreement for categorical data. Biometrics, v. 33, n. 1, p. 159–174, 1977. https://doi.org/10.2307/2529310
LUNDBERG, S. M.; LEE, S. I. A unified approach to interpreting model predictions. In: Advances in Neural Information Processing Systems (NeurIPS 2017), v. 30, p. 4765–4774, 2017.
LUNDBERG, S. M. et al. From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, v. 2, n. 1, p. 56–67, 2020. https://doi.org/10.1038/s42256-019-0138-9
McKINSEY GLOBAL INSTITUTE. The economic potential of generative AI: The next productivity frontier. New York: McKinsey & Company, 2023.
MITCHELL, M. et al. Model cards for model reporting. In: Proceedings of the ACM FAccT ‘19, p. 220–229, 2019. https://doi.org/10.1145/3287560.3287596
NATIONAL INSTITUTE OF STANDARDS AND TECHNOLOGY (NIST). AI risk management framework (AI RMF 1.0). Washington, D.C.: U.S. Department of Commerce, 2023. https://doi.org/10.6028/NIST.AI.100-1
OFFICE OF THE COMPTROLLER OF THE CURRENCY (OCC). Model risk management — Comptroller’s handbook. Washington, D.C.: OCC, 2021.
PAGE, M. J. et al. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ, v. 372, n71, 2021. https://doi.org/10.1136/bmj.n71
PAREJA, A. et al. EvolveGCN: Evolving graph convolutional networks for dynamic graphs. In: Proceedings of the AAAI Conference on Artificial Intelligence, v. 34, n. 4, p. 5363–5370, 2020. https://doi.org/10.1609/aaai.v34i04.5984
PARK, J.; KIM, H.; LEE, S. FinRegLLM: A large language model for automated regulatory compliance mapping in financial institutions. IEEE Transactions on Neural Networks and Learning Systems, Early Access, 2024. https://doi.org/10.1109/TNNLS.2024.3381947
RIBEIRO, M. T.; SINGH, S.; GUESTRIN, C. ‘Why should I trust you?’ Explaining the predictions of any classifier. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, p. 1135–1144, 2016. https://doi.org/10.1145/2939672.2939778
RUDIN, C. et al. Interpretable machine learning: Fundamental principles and 10 grand challenges. Statistics Surveys, v. 16, p. 1–85, 2022. https://doi.org/10.1214/21-SS133
TAPIADOR, J.; HERNANDEZ, R.; MORALES, M. TEVV-F: Extending the NIST AI RMF test-evaluate-verify-validate model for regulated financial environments. Journal of Banking Regulation, v. 24, n. 3, p. 289–308, 2023. https://doi.org/10.1057/s41261-022-00198-8
THOMAS, J.; HARDEN, A. Methods for the thematic synthesis of qualitative research in systematic reviews. BMC Medical Research Methodology, v. 8, p. 45, 2008. https://doi.org/10.1186/1471-2288-8-45
WEBER, M. et al. Anti-money laundering in Bitcoin: Experimenting with graph convolutional networks for financial forensics. In: KDD 2019 Workshop on Anomaly Detection in Finance. arXiv:1908.02591, 2019.
XIAO, H.; RASUL, K.; VOLLGRAF, R. GAN-based adversarial attacks on AML transaction monitoring systems: Implications for BSA compliance. Journal of Financial Crime, v. 29, n. 4, p. 1312–1329, 2022. https://doi.org/10.1108/JFC-11-2021-0244