Could GenAI expose CFOs to systemic vulnerabilities?

Financial institutions are accelerating their adoption of artificial intelligence (AI), driven by advancements in generative AI (GenAI) and large language models (LLMs). These technologies promise transformative efficiencies, from automating regulatory compliance to optimizing financial operations. However, a closer examination reveals risks that could destabilize the systems underpinning modern finance.

Recent findings from the Financial Stability Board highlight a growing concern: the reliance on AI introduces vulnerabilities that extend beyond individual institutions. These vulnerabilities, if left unaddressed, could cascade through the financial system, amplifying disruptions in already volatile markets.

AI has been a mainstay in finance for years, powering credit assessments, fraud detection, and operational optimization. The latest developments in GenAI and LLMs are expanding its scope, enabling applications such as document summarization, code generation, and market sentiment analysis. Investment in AI across financial services is expected to more than double by 2027, with spending projected to reach $400 billion.

While the benefits are evident, the infrastructure enabling AI has become highly specialized. Financial institutions increasingly depend on external service providers for hardware, cloud services, and pre-trained models. These dependencies have concentrated critical functions in the hands of a few technology firms, creating operational vulnerabilities. A disruption to any of these providers could affect essential systems, from payments to risk management.

Concentration and Correlation Risks

The growing reliance on common AI models and data sources raises another concern. Market participants adopting similar tools and algorithms could inadvertently create synchronized behaviors. For example, an error in a widely used model could lead to uniform mispricing or misjudged credit risk across institutions. During periods of market stress, such correlations could exacerbate volatility, intensifying liquidity crunches and price dislocations.

This risk is magnified by the rise of pre-trained AI models. Many financial institutions lack the resources to develop their own systems and instead rely on pre-built solutions. While this approach lowers the barrier to AI adoption, it also reduces diversity in methodologies, increasing the likelihood of systemic repercussions when faults occur.

Cybersecurity and Governance Challenges

AI has not only increased the complexity of financial operations but also widened the attack surface for cyber threats. Generative AI tools can be exploited by malicious actors to create sophisticated phishing campaigns or generate fraudulent data. Simultaneously, the reliance on external cloud providers introduces exposure to cyber events affecting these third-party vendors.

Governance remains another significant challenge. Many financial institutions are still grappling with the opacity of AI systems. Models trained on unstructured or proprietary data often lack transparency, making it difficult to assess their reliability or bias. As these tools are integrated into core financial operations, ensuring they produce consistent and accurate outputs becomes critical.

Systemic Implications for Financial Stability

The risks associated with AI adoption are not confined to individual firms. As AI becomes more central to financial services, its vulnerabilities have the potential to affect the broader system. Concentration among a limited number of AI service providers, combined with market-wide correlations in model behaviour, creates the conditions for systemic risk.

If a key provider of AI infrastructure suffers a failure, multiple institutions could face simultaneous disruptions. This would not only impede their operations but also diminish confidence in the financial system as a whole. The stakes are particularly high for large, interconnected firms whose stability is essential to market functioning.

Addressing the Risks

To mitigate these risks, financial institutions need to prioritize robust governance frameworks. Effective oversight requires dedicated AI governance committees that can evaluate model performance, ensure compliance, and address emerging risks. Institutions must also demand greater transparency from AI vendors, particularly regarding training data and system architecture.

Regulators have a role to play as well. Existing frameworks must be adapted to address the unique challenges posed by AI. This includes international collaboration to monitor cross-border dependencies and the establishment of standards for AI governance and risk management.