87% of the 23 financial professionals the Fed surveyed said AI could set off a shock to the financial system in the next 12 to 18 months, up from 50% in the spring.
The Federal Reserve reported that 87% of the financial professionals it surveyed said AI could set off a shock to the U.S. financial system in the next 12 to 18 months. No other risk was named as often. In the Fed’s spring report, 50% named AI, and it ranked third behind geopolitical risks and an oil shock.
The survey is part of the Financial Stability Report. The Fed publishes that report twice a year to describe threats to the U.S. financial system.
The respondents described three AI risks: a market correction, cyberattacks, and the debt and complex financing behind AI. The rest of the report is the Fed’s analysis. In that analysis, the Fed gave its own assessment of all three risks.
Key takeaways
- 87% of the 23 financial professionals surveyed said AI could set off a shock to the financial system, up from 50% in the Spring survey.
- 52% said a market correction could do the same, up from 35%.
- 48% pointed to higher long-term interest rates, up from 20%.
- 43% cited cyberattacks, up from 20%.
- 43% worried that U.S. government debt is not sustainable, up from 30%.
Fed says AI creates new security challenges
Many respondents said AI is making sophisticated cyberattacks easier. Others pointed to “damage caused by frontier AI models themselves.” They viewed cyberattacks as one of the two risks with the most severe consequences.
The Fed gave its own view of cyberattacks in its analysis. It said AI models and AI agents recently got better at finding and exploiting security flaws. Those advances created new security challenges for financial institutions, financial market infrastructure (i.e. the systems that process trades and payments), and third-party service providers. Reports of AI agents being involved in cyberattacks and other security incidents also increased since the previous Fed report in May.
The Fed then explained how the damage could spread. A cyberattack or other security incident that limits the services of one financial institution could reduce market liquidity, shake investor and depositor confidence, and force asset sales. An attack on a critical third-party service provider could affect many institutions at once. The damage would likely be worse if those institutions could not easily replace the provider.
Companies borrowed $680 billion for AI data centers
The Fed estimated that companies issued more than $680 billion in debt for AI data centers through October 2026. In all of 2023, the total was $130 billion. Investment-grade technology companies issued most of the 2026 debt.
Debt for AI data centers made up about 31% of all investment-grade debt issued by nonfinancial companies in 2026. In 2023, the share was under 6%.
Respondents said that heavy debt could make losses larger if AI-related stocks fell sharply.
The Fed also added up payments that large technology companies have promised to make in the future. Those promises do not show up as debt on a balance sheet. The Fed identified seven companies. Five are hyperscalers: companies that run the biggest data centers. These are Amazon, Alphabet, Meta, Microsoft, and Oracle. The other two are the chip makers Broadcom and Nvidia.
Those seven companies committed to more than $1 trillion in leases that have not started. They also made more than $2 trillion in other commitments, such as guarantees and purchase contracts.
The Fed explained how these commitments help other companies borrow. A hyperscaler often signs a long lease for a data center before a developer builds it. The developer then uses the lease to secure construction financing. Chip makers often guarantee lenders a minimum future value for chips that a buyer pledges as collateral.
The Fed said these commitments tie the largest technology companies to smaller, riskier firms. These include the developers that build data centers.
In a downturn, the commitments could leave the largest companies taking the losses themselves. That would keep the damage from spreading through the financial system. It also means the largest companies have far more at risk than their balance sheets show.
The Fed said the largest U.S. banks had close to $1 trillion of exposure to four groups:
- large technology companies
- data center developers
- special-purpose vehicles and private credit funds tied to AI data centers
- utilities
Each year, the Fed runs a stress test on the largest banks. It checks whether each bank could survive a severe downturn across all of its lending and trading. The stress test analyzes credit card loans, mortgages, business loans, commercial real estate loans, and trading, including the money banks have tied up in the four groups above.
More broadly, the Fed said the links among technology companies, lenders, and special-purpose vehicles created vulnerabilities that required close monitoring.
Methodology
Staff at the Federal Reserve Bank of New York ran the survey in September and October 2026. Respondents work at broker-dealers, banks, investment funds, and advisory firms. They were asked about risks to the financial system over the next 12 to 18 months.

