Generative Artificial Intelligence (GenAI) offers financial institutions with new opportunities to provide better customer experience, enhance risk management, compliance reporting, and improve efficiency. However, the deployment of GenAI in the financial sector requires the industry to recognise and mitigate potential risks comprehensively. The International Monetary Fund (IMF) has published a Fintech Note exploring the unique risks presented by GenAI and proposing ways to address them.

Risk considerations

Data privacy

GenAI poses several privacy concerns. These include data leakages from the training data sets, the disclosure of anonymised data through inferences, and AI/ML “remembering” information about individuals even after use. To address these privacy concerns, financial institutions should use differential privacy to preserve private information.

Explainability

Explainability of decisions and actions taken by AI algorithms in financial institutions is a complex issue. AI algorithms have a dense architecture that relies on numerous parameters. The input signals might be difficult to interpret, and the output might contain biases. To address these problems, financial institutions should employ transparent models that facilitate interpretability through proper documentation, visualisations, and code explanations.

Synthetic data

Generating synthetic data refers to the process of applying algorithms to produce data that resembles real-world data. Synthetic data facilitate the anonymisation of sensitive information but pose a risk as the synthetic data might contain synthetic biases. Financial institutions should remain vigilant of the potential risks that synthetic data might pose when used to train AI models and consider testing the data with a real-world perspective to ensure they are producing trustworthy results.

New cybersecurity threats

The deployment of GenAI in the financial sector could lead to new cybersecurity threats, including adversarial attacks, where cyber attackers exploit vulnerabilities in the models to gain access to the models and manipulate their outputs. Financial institutions should carefully evaluate cybersecurity risks associated with GenAI deployment and develop comprehensive defence strategies.

Regulatory and operational risks

The use of GenAI in the financial sector could lead to operational and regulatory risks such as business continuity disruptions, compliance risks, and ethical concerns. Financial institutions need to adopt a risk-based approach and integrate GenAI governance in regulatory and risk frameworks.

Recommendations

• Financial institutions should use differential privacy to preserve private information.
• Employ transparent models that facilitate interpretability through proper documentation, visualisations, and code explanations.
• Consider testing data with a real-world perspective to ensure they are producing trustworthy results.
• Carefully evaluate cybersecurity risks associated with GenAI deployment and develop comprehensive defence strategies.
• Adopt a risk-based approach and integrate GenAI governance in regulatory and risk frameworks.