Federal Reserve Governor Michael Barr said there needs to be clear guardrails to prevent risks as the financial sector looks to adopt artificial intelligence in its core functions.

Regulators need to get the balance right between innovation and stability to ensure that AI boosts growth and productivity over the long-term, Barr said at the Singapore FinTech Festival on Wednesday.

“Speaking only about the United States, I worry that we are going to let that pendulum swing too far and lower our guardrails in a way that opens us up to too much risk,” he said in a panel with several other central bankers.

The financial sector is moving fast to deploy AI across their businesses, such as in customer service, document summarization, sales and marketing, and public relations, he said. “But firms are also beginning to explore how generative AI might fit into their core functions. And that’s an area that requires great care.”

Policymakers should ensure the technology doesn’t introduce risks from market manipulation and collusion, according to Barr. He flagged a scenario where AI systems trade with one another in a way that increases volatility or even introduces systemic risk.

Regulators should also be wary that AI could be trained on skewed data or techniques that could introduce new biases into the financial system, he said.

Read More: AI Is Driving US Growth With Economy Plunged in Data Blackout

Fed officials elected to cut their benchmark interest rate at each of their last two policy meetings following a sharp slowdown in hiring over the summer. Recent public comments indicate they are divided over the need for a third reduction in December, though investors are betting on one, according to futures.

Although AI will transform economies, there are a range of outcomes as to how that will play out, the Fed governor said. In the first, adoption of generative AI could augment existing tasks and roles. In the second, it could lead to a transformative impact, where work and leisure undergo radical change that boosts efficiency and remakes firms with new business models.

“These different kinds of scenarios and lots of intermediate steps in between are possible, and we need to track them,” Barr said.

In his prepared remarks, Barr pointed to a New York Fed survey showing AI has led employers to scale back hiring plans — a development the Fed governor suggested may be contributing to slower levels of job creation — though he didn’t comment on the near-term outlook for monetary policy.

He also pointed to the possibility that the trillions of dollars in planned capital investment into data centers could drive significant economic change, including productivity gains.

“Investment in capital generally raises labor productivity and offers the potential for higher output growth without pressure on inflation over the longer term,” Barr said. “As I have discussed in previous remarks, if these changes are significant, they can also affect the conduct of monetary policy.”