← News

Fed's Michael Barr lists the ways AI could rattle the financial system

Federal Reserve Governor Michael Barr said Monday that AI could bring new risks to trading, cybersecurity and consumer protection, building on a Fed report last week that found AI is now the most-cited worry about financial stability.

When a Fed governor starts counting the AI industry's IOUs and asking who authorized your bank's AI agent, the boom has become a financial-plumbing question. It also lands the same week as Nvidia earnings and a record Anthropic IPO.

Federal Reserve Governor Michael Barr spoke Monday, Nov. 16, at the Seventh Annual ACM International Conference on AI in Finance in Milan. The views are his own, not necessarily the Board's. He opened by calling himself an optimist about AI's long-term promise. Those lines are his, in the speech the Fed posted.

He built on the Fed's Financial Stability Report, released Friday. Reuters reported that 87% of the contacts surveyed for that report cited AI as a salient risk to the financial system. Salient, here, means a risk they actually named, not a side note. The 87% figure is Reuters' account of the report.

The debt on the books is the smaller number. On-balance-sheet debt of the seven largest AI firms has grown fast in 2026 and is expected to exceed $800 billion before year-end. On the balance sheet means debt the company reports as its own. Barr compared that pile with telecom debt in the late-1990s broadband boom. The $800 billion and the comparison are from his speech.

Off the books is larger. The report estimates that the potential maximum nominal value of long-term contracts, leases, and residual value guarantees now exceeds $3 trillion. Nominal value is the face amount of those promises, not a price someone would pay today. A residual value guarantee is a promise to cover the gap if equipment is worth less than expected when a lease ends. Barr said that complexity can be opaque, hard to see through, and can make a selloff worse because it is hard to know where the risk sits. The $3 trillion estimate is the report's. The warning about opacity is his.

Then the loop. Hyperscalers and big GPU makers hold over $100 billion in equity stakes in AI labs and cloud startups that then spend the money renting or buying compute from those same investors. A hyperscaler is one of the giant cloud companies. A GPU is the graphics chip used to train and run AI models. An equity stake is an ownership share. Eventually those smaller firms have to replace supplier money with revenue from paying customers. That account is Barr's.

Trading and agents, as Reuters summarized the speech. AI trading models could herd, meaning crowd into the same bets at the same time, and disrupt markets. Self-learning algorithms could manipulate markets. Banks may be unable to confirm whether an AI agent acting in a retail account is actually authorized to do it. An agent, here, is software that can take actions in an account, not just draft a reply. Fast-moving legitimate agents could accelerate a bank run, a rush of withdrawals that can bring a bank down. He also flagged AI-enabled cyberattacks and rogue agents. Those risks are Reuters' summary of what he said.

His closing standard, in his words: decisions based on these models must be "well controlled, numerically and legally precise, explainable, and replicable," and "AI developers still struggle to some extent with all of those criteria." Explainable means a person can say why the model did what it did. Replicable means someone else can get the same result again.

The picture is Federal Reserve Governor Michael Barr speaking at a podium at the Peterson Institute for International Economics in Washington. It is a Federal Reserve photograph from September 2025. It is a file photo, not a picture from Monday's speech in Milan. The credit is Federal Reserve photo (public domain). Public domain means anyone can reuse the photo.

In plain terms, a Fed governor who says he is optimistic about AI spent a Monday in Milan listing how it could shake trading, cyber defense, and consumer accounts, and how the industry's promises, on the books and off them, have gotten very large. The $800 billion is debt the biggest AI firms are expected to carry by year-end. The $3 trillion is the report's ceiling on contracts, leases, and guarantees that do not sit on those balance sheets. He also described a loop in which the companies selling compute own stakes in the labs that rent it back. None of this is a new rule. It is his read, and he said it is not necessarily the Board's.

RELATED

ONLINE…

Comments

guidelines

Loading…

Loading…

Sources