Regulations on AI in banking should focus on outcomes, not models

Artificial intelligence is revolutionizing the financial services sector, from automating credit assessments to enhancing fraud detection. However, as AI’s footprint expands, so does the debate over how to regulate its use effectively. I felt this yin and yang firsthand when I headed up AI innovation at Chase. While some advocate for stringent oversight of AI models and infrastructure, a more consumer-centric approach — leveraging existing financial regulations — may offer a balanced path that safeguards consumers without stifling innovation.

Current regulatory efforts often zero in on the technical aspects of AI, emphasizing transparency, bias mitigation and rigorous oversight. For instance, the European Union’s proposed AI Act seeks to categorize AI applications based on risk levels, imposing strict requirements on high-risk systems including those in financial services. Similarly, in the United States, discussions around AI regulation have gained and then lost momentum at the Federal level, with various states contemplating laws to oversee AI’s deployment in sectors like health care and finance. For example, California leads the way with new AI laws promoting transparency, privacy and ethical practices across various industries.

However, the financial industry is already one of the most regulated sectors globally. Regulations such as the General Data Protection Regulation, or GDPR; the California Consumer Privacy Act, or CCPA; and the Dodd-Frank Act impose stringent requirements on data protection, consumer rights and financial stability. Layering additional AI-specific regulations onto this framework could lead to redundancy, increased compliance costs and potential roadblocks — not to mention confusion.

Am I anti-regulation for AI? Not at all. I’m suggesting that shifting the regulatory focus from the intricacies of AI models to the outcomes they produce offers a pragmatic alternative. This approach emphasizes that AI-driven decisions should align with existing consumer protection laws, ensuring fairness, transparency and accountability.

Rather than dissecting the algorithms themselves, regulators can assess whether AI-driven decisions comply with established consumer protection standards. For example, if an AI system denies a loan application, the decision should adhere to fair lending practices, providing clear reasons for the denial and ensuring no discriminatory factors influenced the outcome.

Existing regulations already mandate fairness and accountability in financial services. The Equal Credit Opportunity Act prohibits discrimination in credit transactions, regardless of whether decisions are made by humans or AI systems. Next, the Consumer Financial Protection Bureau oversees financial institutions to ensure compliance with consumer protection laws, extending its purview to AI-driven practices. Additionally, the GDPR and CCPA mandate data privacy and grant consumers rights over their personal information, affecting how AI systems handle and process data.

Existing financial regulations can be effectively applied to govern AI applications. This includes anti-discrimination and fairness where financial institutions must ensure that AI models used in credit scoring or mortgage approvals do not perpetuate biases. Regular audits and bias testing can help maintain compliance with anti-discrimination laws. In addition to this, fraud prevention and AI transparency are equally important.

AI systems play a pivotal role in detecting fraudulent activities. However, their deployment must align with anti-money-laundering and know-your-customer regulations, ensuring that AI-driven alerts and decisions are transparent and justifiable. Lastly, market stability and AI in trading are crucial considerations, given that AI-driven trading algorithms must operate within the boundaries set by financial regulators like the Securities and Exchange Commission and the Commodity Futures Trading Commission. These bodies ensure that such technologies do not destabilize markets or engage in manipulative practices.

While regulation is essential, over-regulating AI models and infrastructure can have unintended consequences. Excessive regulatory requirements can lead to increased compliance costs, as financial institutions may face substantial expenses to meet overlapping regulatory demands, diverting resources away from innovation. Stringent regulations can slow down innovation by deterring the development and adoption of beneficial AI technologies, ultimately limiting advancements that could enhance consumer experiences.

An alternative approach is adopting a risk-based, outcomes-oriented framework, which allows for flexibility and proportionality in regulation. Principles-based regulation focuses on ethical AI use and desired outcomes rather than prescribing specific technical measures, allowing institutions to adapt practices to their unique contexts. Regulatory sandboxes provide a controlled environment for financial firms to test AI innovations under regulatory supervision, fostering experimentation while safeguarding consumers.

As AI continues to reshape financial services, regulators face the challenge of protecting consumers without hindering technological progress. We all want this balance, and protection is good. However, by focusing on the outcomes of AI applications and enforcing existing consumer protection laws, policymakers can ensure that AI serves the public interest. This approach not only safeguards consumers but also encourages responsible innovation, allowing the financial sector to evolve while maintaining trust and integrity.

In essence, the goal should be to create a regulatory environment where AI can thrive responsibly, delivering benefits to consumers without compromising on ethical standards or stifling technological advancement.