BNY Integrates AI into Financial Operations, Partnering with OpenAI

In early 2023, following the release of ChatGPT by OpenAI, BNY convened its senior executives to discuss the integration of artificial intelligence into its financial operations. While AI has sparked a competitive race across various sectors, including healthcare and law firms, banks like BNY must proceed with caution. The potential risks, such as rogue agents or hallucinations, could trigger a financial meltdown. Additionally, regulatory compliance and the protection of sensitive data and personally identifiable information add layers of complexity.

Despite these challenges, BNY's executives recognized the transformative potential of AI and its significance for the company's future. Founded in 1784 by Alexander Hamilton, BNY is the oldest bank in the country. Sarthak Pattanaik, appointed to lead BNY's AI efforts, emphasized the bank's history of innovation, stating, "You don’t get to 240 years without being innovative."

BNY swiftly incorporated AI into its infrastructure, launching Eliza, an AI tool named after Hamilton’s wife. Powered by models like OpenAI’s GPT-4 and Google’s Gemini, Eliza enables employees to build AI agents for tasks such as compliance and advanced reasoning. Recently, BNY announced a partnership with OpenAI to collaborate on financial services use cases.

New AI leaders at major financial institutions, like Pattanaik, face the decision of developing products in-house or seeking outside vendors. Traditionally, banks have built new technology internally, but this approach is evolving. David Haber, a veteran of fintech startups and Goldman Sachs, now at Andreessen Horowitz, notes a cultural shift towards adopting best-of-breed technology. Some banks, including BNY, are training their own open-source systems and relying on proprietary models from companies like OpenAI.

The cloud revolution of the 2010s, which saw financial institutions partnering with third-party providers like Amazon Web Services, has accelerated this trend. Banks are increasingly recognizing the benefits of adopting cutting-edge technology.

Banks have long used machine learning for tasks like consumer behavior analysis and underwriting. However, generative AI models from companies like OpenAI and Anthropic have disrupted classical machine learning. These models can be trained for a wider variety of tasks, simplifying the need for bespoke technologies and solutions.

Banks are turning to external AI systems developed by companies with extensive resources and data. David Griffiths, Citi's chief technology officer, highlights the importance of working closely with model providers to give feedback and stay updated on the rapidly evolving technology.

BNY works with multiple proprietary models, including Anthropic, Google’s Gemini, and OpenAI, as well as open-source options like Meta’s Llama models and Mistral. Pattanaik mentioned that BNY's partnership with OpenAI involves intellectual capital sharing beyond just computing resources.

Many banks have developed virtual assistants like BNY’s Eliza and Citi Assist, which help employees with tasks such as coding and procedural questions. Griffiths notes that AI currently serves an assistive purpose, enhancing productivity. However, the future of AI agents, which can take actions on their own, poses both opportunities and risks, including the potential for hallucinations.

The presence of sensitive information in finance limits the types of AI use cases banks pursue. BNY, operating mostly in the institutional space, has more freedom but still avoids training models with personal information. When necessary, the bank uses encryption and red team testing to simulate cyberattacks.

Banks are adopting a more cautious approach when purchasing third-party AI tools, implementing a separate procurement process to ensure compliance and security. Lindsay Fitzgerald, a former banker and founder of Vesey Ventures, notes that banks are more likely to buy software that doesn’t touch their core infrastructure, such as AI tools for accounts receivable collection.

In just a few years since the launch of ChatGPT, AI has already transformed banking. A 2024 report ranked major global banks by AI adoption, with all 50 analyzed referencing AI in investor relations documents. Over half publicly report on use cases in production. The leading bank in the rankings valued its AI use cases at around $2 billion.

Griffiths acknowledges the competitive pressure but emphasizes Citi's focus on using AI to make or save money. The bank has clear plans for the next year, though the rapid evolution of AI models makes future capabilities uncertain.