In early 2023, shortly after Openai released Chatgpt to the world, BNY gathered its senior executives to decide how to incorporate artificial intelligence into its financial empire.
While AI has established an arms race across sectors from health care to law firms, banks like BNY still have to practise caution. After all, a rogue broker or hallucination could trigger the next financial crash, not to mention the process of traditional Chinese tape and regulations that institutions must consider in terms of sensitive data and personally identifiable information of clients.
Still, BNY executives recognize that AI is probably one of the most important technological developments in the company’s history and they don’t want to miss the ship. Originally founded by Alexander Hamilton in 1784, BNY is the oldest bank in the country. Just as Sarthak Pattanaik, who chose to lead its AI efforts a few months later, wealth: “You won’t reach 240 years without innovation.”
BNY took quick action to fold AI into its infrastructure, launching a tool called Eliza last year, named after Hamilton’s wife, powered by existing models such as OpenAI’s GPT-4 and Google’s Gemini. Eliza allows employees put up AI agents (such as chatbots) provide niche topic expertise in areas such as compliance and more advanced inference tasks. Just a few months ago, BNY announced a new partnership with OpenAI, which will collaborate in financial services use cases.
New AI leaders such as the country’s leading financial institutions, such as Pattanaik, have to weigh whether to develop products rather than searching for external suppliers – working carefully while looking at their shoulders in the well-known competitive and cruel sectors.
Bank norms have long been building new technologies internally. Goldman Sachs even built its own proprietary email client. But that approach is shifting, says David Haber, a veteran of fintech startups and Goldman Sachs, who now serves as general partner at Andreessen Horowitz. “The culture is, if it wasn’t built here, we wouldn’t be interested,” he recalls. “In my opinion, this culture has really started to change dramatically in the last five to six years.” Some banks, such as BNY, are training their own open source systems, relying on like Yuan Or Mistral as a starting point and trained based on its own data and move to proprietary models in certain use cases, such as those from OpenAI.
Cloud Revolution, When Financial Institutions Begin cooperate Like third-party providers Amazon Web services in the 2010s helped stimulate the transition. Huber told wealth. “I suspect AI will only accelerate this trend.”
Banks are no strangers to artificial intelligence and use machine learning Decades Analyze consumer behavior and perform core functions such as underwriting. But Citi’s chief technology officer David Griffiths told wealth Generative AI models advocated by OpenAI and Anthropic companies undermine many of the classical machine learning that Banks has previously worked with data scientists and researchers.
If machine learning is designed for specific use cases, such as fraud modeling or document recognition, large language models can be trained for a wide variety of tasks. “In some areas, this can actually be a huge simplification force for us because we can get rid of these custom technologies and solutions and suppliers and use something more versatile,” Griffith said.
When it comes to these models, banks mostly seek help outside the wall: large-scale AI systems developed by in-depth companies like Openai and trained on trillions of data points. “We want to be able to work as closely as possible with the model provider itself, because the technology is so new, we want to be able to provide feedback,” Griffith said.
Although BNY publicly announced a partnership with OpenAI, Pattanaik said the bank works with all three proprietary models (Human, Google’s Gemini and OpenAI), as well as open source options including Meta’s Llama Models and Mistral, which is managed for security purposes. He declined to detail how BNY and Openai cooperate with other companies, saying the two companies work more closely on “intellectual capital sharing” rather than just paying for computing resources.
Many banks have developed their own virtual assistants, such as BNY’s Eliza and the company’s Citi Assistant, roll out There were about 150,000 employees at the end of last year. As the tool’s name suggests, Griffiths said AI is now aimed primarily at auxiliary purposes, making employees more “productive” by helping code or answering procedural questions about bank bureaucracy. “As agent models become real and unfold, it will be interesting to see how this might change the shape of the workforce,” Griffith said. “The next six to 12 months will be really told throughout the industry.”
AI agents describe artificial intelligence systems that can act on their own, rather than simply asking people to act or generating text or images, which also creates increased risks and potential problems such as hallucinations. All generated AI models can suffer from “illusion” – AI models confidently provide inaccurate output, but if these inaccurate people now lead to financial transactions, the consequences may be even worse. Griffiths said quality control, such as fine-tuning of a model and providing specific data, can help reduce the possibility of concatenation.
The existence of sensitive information across finance will naturally limit the types of use cases banks are pursuing, at least so far. Since BNY operates primarily in the institutional sector, this means it does not hold consumer data such as credit card or mortgage information, it has more freedom than most competitors. Even so, Pattanaik said BNY tried to avoid using personal information training models. If so, the bank will adopt a “walled garden” approach that simulates cyberattacks through encryption and red team testing.
Lindsay Fitzgerald says Morgan Stanley He then led the corporate venture capital division of American Express and then started his own company Vesey Ventures. Fitzgerald, who sees major banks as supporters, said many banks have built a separate, more tedious procurement process for purchasing third-party AI tools. “There is a flag, does this use AI?” she told wealth. “If you hit that flag, you’ll sign up for a few weeks.”
As a result, banks tend to buy software that doesn’t touch its core infrastructure. Fitzgerald highlights Stuut, a portfolio company that helps collect payments through AI, including voice agents. “This is an extremely low risk implementation of AI,” she said, because it does not touch on the personal information of customers. The application layer of startups catering to financial institutions is only possible to grow.
AI has transformed banking in just a few years since Chatgpt’s launch. according to 2024 Report From the research and analysis platform, obvious insights will be the ranking of major global banks adopted by AI, and the company analyzed all 50 in at least one investor relations document. More than half of the publicly reported use cases in production. The leading banks in the ranking, JPMorgan Chaseannounced at a meeting in September that AI use cases within the bank were about $2 billion.
Griffiths said Citi, which is clearly ranked eighth, is aware of the pressure of competition, although the bank is trying to focus on making money or saving money using AI. “We are working to avoid reactions to what we read or hear about what other people are doing,” he said. “For next year or so, we have very clear plans, but of course, who knows the capabilities of these models in six months.”
