Barr warns AI financing and autonomous agents raise stability risks
FED Speech

Barr warns AI financing and autonomous agents raise stability risks

Federal Reserve Governor Michael S. Barr warned that AI financing, trading systems and autonomous agents pose financial stability risks. Speaking at an AI finance conference in Milan, he called for stronger controls while emphasizing AI's potential benefits.

Debt beyond the balance sheet

Barr cited the Federal Reserve's Financial Stability Report: on-balance-sheet debt at the seven largest AI firms is projected to exceed $800 billion by end-2026.

Potential maximum nominal off-balance-sheet obligations exceed $3 trillion, including leases, energy contracts and collateral guarantees.

These undiscounted commitments are not directly comparable with discounted balance-sheet debt.

Hyperscalers and chipmakers also hold over $100 billion in equity investments in AI labs and cloud start-ups that buy their services.

Barr said this circular funding must eventually give way to customer revenue.

He warned that opaque obligations, hedge fund leverage and dealer counterparty exposures can amplify repricing, despite large banks' capital buffers.

Shared models multiply vulnerabilities

Generative AI can broaden access to trading tools and improve surveillance, Barr said, but autonomous trading is not yet deployed at large scale.

Similar models and training data risk synchronized trades and shared failures.

Research identifies risks of autonomous collusion, spoofing and bubble exploitation, while costly systems can concentrate market power.

AI also helps defenders patch cyber vulnerabilities faster, but attackers exploit shared providers and previously inaccessible weaknesses.

Barr called for cooperation to strengthen cyber defense.

Guardrails before autonomous action

Barr said AI financial advice raises conflicts of interest and lending models risk discrimination.

He called for explainability, fairness monitoring, model validation and accessible disputes.

Banks also need authentication and explicit consent controls as autonomous agents risk unauthorized transactions and faster deposit runs.

Report an error