- Key insight: The NYSE has begun using Claude agents as autonomous engineering collaborators.
- What's at stake: Regulatory, resilience and market integrity risks if probabilistic AI lacks continuous oversight.
- Supporting data: NYSE processes more than a trillion messages on peak trading days.
Source: Bullets generated by AI with editorial review
NEW YORK — The highly regulated New York Stock Exchange, founded in 1817, is moving quickly with agentic AI projects, using Anthropic's Claude generative and agentic AI extensively throughout the organization.
"Eighteen months ago, it was more of a chat interface, and people were used to using it for code completion," said Sridhar Masam, chief technology officer at the exchange, during a virtual event Anthropic hosted on Tuesday. "Now with its agent ticket reasoning capabilities, it's more independent. It's more of a collaborator than an assistant. That was a big fundamental shift in how people use AI."
Masam was referring to Anthropic agents based on Claude 3.7 Sonnet and the Claude Agent SDK that use "chain-of-thought" reasoning to process complex, multistep tasks like customer service, IT support or software bug resolution.
"We see AI as a tremendous accelerator in 2026 as adoption internally grows and as we move on from experimentation to production to scale," Masam said.
The stock exchange's work with Anthropic mirrors AI projects large financial institutions like JPMorganChase and Goldman have been undergoing.
"These players are moving from AI as a point solution in a single step in a workflow to embedding AI into core applications, such as digital banking platforms, payments processing systems, credit underwriting engines and fraud detection platforms," said Alenka Grealish, lead analyst for emerging tech intelligence at Celent. "While most [financial institutions] are still working on fixing operations, early movers are increasingly leveraging AI to reinvent the institution."
Steve Rubinow, former chief information officer at the New York Stock Exchange and current associate teaching professor at the Illinois Institute of Technology, was not surprised to hear of the stock exchange's collaboration with Anthropic.
"Everybody's got to be experimenting," Rubinow told American Banker. "Nobody can afford to be a late adopter, it doesn't matter what industry you're in, because the technology is moving so quickly, and people see there's a competitive advantage or competitive necessity. The question is, what experiments do you choose? Where are you in the risk reward curve?"
Financial services organizations need to be more conservative than those in less regulated industries, "but it sounds like they have the right tone," Rubinow said.
Where Claude is helpful
The NYSE has been reengineering its development process and using Claude in coding, writing tests and documenting new code, Masam said.
Where software development used to be a matter of "buy what you can, build what you must," he said, now it's becoming a matter of combining multiple models, multiple vendors, platforms, data and internal capability.
"Assembly is becoming a key aspect where you need to assemble all these things to have a solution," Masam said.
His team also used Claude code to build a reference implementation for the
"We see AI being part of our journey to accelerate that implementation," he said.
Claude models are effective in processing large documents and applying rules, Masam said. The NYSE has used the technology to build agents that review proxy filings, audit Securities and Exchange Commission filings and generate news classifications, he said.
Need for strong governance
The NYSE processes more than a trillion messages on a peak trading day.
"Even at that scale, system resilience, determinism are paramount," Masam said.
Using AI to develop software brings more accountability, Masam said.
"Traditionally, we are so used to building deterministic platforms," he said. "You write code requirements and build. Now, with AI being probabilistic, the accountability doesn't end when the project goes live, but on a daily basis, you have to monitor behavior and outcomes."
Rubinow said there are a few things organizations like the NYSE need to keep in mind as they deploy agentic AI.
"What's important in this new era, and some things haven't changed, is data," he said. "You've got to pay attention to the data, because if you don't pay attention to the data, it doesn't matter how beautiful the software you're passing it through is, we know what you're going to get out."
He also advised people deploying AI to think more like conductors than coders.
"You have to be a systems thinker," Rubinow said. "You can't just look at little pieces of software and glue them all together and say, 'OK, now I have a system.' You have to take a step back and look at the entirety, because you can't peek under the cover of a large language model and say, 'Oh, I can see all the gears moving. I know how it works.' You have to focus on the things you can control, like the good old-fashioned algorithms and code and the large language model components, whether it's one piece or it's a series of agents, and take a step back and look how the whole system performs, not just individual pieces."
And as always, there has to be a human in the loop, he said.
"You've got to scrutinize your results more than ever, because you want to make sure that what you are using and what you're passing along is indeed acceptable for whatever application you're using," Rubinow said. "And sprinkle in it a healthy dose of safety and ethics, because that's all part of the human input as well."
