Federal Reserve stands up generative AI incubator

The Fed is looking to test concepts in areas that consume and generate a lot of data — such as payment rails and supervision and regulation.
Sunayna Tuteja, chief innovation officer of the Federal Reserve System, speaks during a fireside chat at the “Federal Innovation Series: Leading in the Era of AI” event, presented by Microsoft and FedScoop.

While some agencies in the public sector have pumped the breaks on the use of generative artificial intelligence, the Federal Reserve System has set up a generative AI incubator, looking to be “bold” but responsible in its use of the technology, according to the agency’s chief innovator.

The generative AI incubator takes the private sector concept and a bias toward action and marries it with the need to be responsible and manage risk, Sunayna Tuteja, chief innovation officer of the Federal Reserve System, said last week at the “Federal Innovation Series: Leading in the Era of AI” event, presented by Microsoft and FedScoop.

“At the Fed we’re really good at, you know, writing white papers and thinking, and thinking about thinking about thinking more,” Tuteja said. But with generative AI, “this is one of those technologies where you have to think but you also have to do, so how do we create a framework that enables us to do that?”

Tuteja’s comments came just after President Joe Biden signed a landmark executive order on AI and the Office of Management and Budget issued guidance to agencies on the use of the technology and, among other things, evaluate potential applications of generative AI.


Tuteja looked back on her experience in the private sector at TD Ameritrade, prior to joining the Fed in 2021, where incubators are fairly common as inspiration for the idea, she said.

“In the private sector, we used to say: Go from idea to IPO — the Fed doesn’t let you IPO obviously, so for the Fed it is how do we go from idea to implementation,” she said of incubators. “And the reason this construct is helpful is it really enables us to interrogate this technology, but through this eclectic combo of the bold and the boring. In order for us to do big, baller, bold things, it actually requires us to do a lot of important boring things in a very consistent manner.”

While Tuteja didn’t speak to any specific examples of generative AI concepts being tested in the incubator currently, she pointed to areas that consume and generate a lot of data as the best places to start — such as payment rails and supervision and regulation.

“These payment rails consume a lot of data, but they also generate a lot of data. So the way we’re applying generative AI is really as we look at all the data that’s coming from our ACH payment rails or FedNow or cash, the usage statistics, etc., how do we now take this … how do we apply the right models so that we can extract insights that enables us to make decisions not just from … an investment perspective, but also from a customer experience perspective, really understanding which payment rail does the customer want to use, at what time, for what purpose and kind of align our business decisions accordingly,” she said, adding that the goal is to augment and turbocharge the work humans are doing.

The Federal Reserve System’s generative AI incubator is rooted in several guiding principles. Top on that list is “we’ve got to put hands on keys,” Tuteja said, meaning practitioners must actually explore and use the technology. “This is not a theoretical exercise. So how do we create the right guardrails and safe spaces within our organization that enables and empowers as many of our colleagues to put hands on keys and start to think about the value of this technology on the businesses that they lead?” she said.


The agency is also taking an approach of “meaningful optionality” and being driven by business problems and business trends, rather than chasing shiny objects, said Tuteja. Further, she said, “it behooves us to be pragmatic about this … from day one, bring in … legal and compliance and security and risk so you’re designing the solution with those appropriate guardrails in mind, because if you don’t do that, you may share something really cool, but it may never scale. So then what’s the value?”

Finally, there must be a bias for action, Tuteja said. “This is not something that we need to do in years and decades. This is stuff that we’ve got to move at a clip that’s about weeks and months and kind of learn by doing.”

Striking a balance between that action and the need to be safe and responsible is a sweet spot for the federal government,” she said.

Tuteja explained: “I think in the public sector, we sometimes are really good at interrogating the risk of doing something new to the absolute nth degree, which is important. But we don’t spend enough time thinking about what is the risk of not doing something new. So that balanced conversation is critical. For us, that’s why the incubator model is important because we’re not going to go start applying these new capabilities to crown jewels and to, you know, to the systems that are doing really important work day in and day out. But that doesn’t obfuscate the fact that we will never do anything, right?”

Billy Mitchell

Written by Billy Mitchell

Billy Mitchell is Senior Vice President and Executive Editor of Scoop News Group's editorial brands. He oversees operations, strategy and growth of SNG's award-winning tech publications, FedScoop, StateScoop, CyberScoop, EdScoop and DefenseScoop. After earning his journalism degree at Virginia Tech and winning the school's Excellence in Print Journalism award, Billy received his master's degree from New York University in magazine writing while interning at publications like Rolling Stone.

Latest Podcasts