SIAgentNews.com
← All news
DevelopingAgentsPolicy

Fed official warns AI agents could trigger bank runs in minutes

Federal Reserve Governor Michael Barr warns that autonomous software could move deposits at speeds that outpace current banking regulations.

· First reported by Startup Fortune

Isometric diorama: picture for the story 'Fed official warns AI agents could trigger bank runs in minutes'
AI-made isometric diorama for this story. No real people or products are shown.
$42 billionAmount lost by Silicon Valley Bank in one day

The short version

  • AI agents could spot yield gaps or stress signals and move deposits in seconds.
  • The risk stems from many agents running on similar models and reading the same data.
  • Bad actors could exploit these systems by feeding false information to trigger mass moves.
  • JPMorgan has already blocked outside AI agents from customer accounts to mitigate risk.

Why it matters

This highlights a shift where financial stability depends on software behavior rather than human psychology. For users, it means your automated savings tools could become part of a systemic risk if they all react to the same news at once.

The speed of automated flight

Federal Reserve Governor Michael Barr recently warned that autonomous AI agents could significantly accelerate the pace of bank runs. Unlike human-driven panics, these software programs can move customer deposits automatically in response to specific market signals or interest-rate changes. This creates a scenario where money could leave a bank in minutes rather than days.

Barr noted that current liquidity management systems and regulatory timelines are not built for this level of speed. He compared the potential for AI-driven runs to the collapse of Silicon Valley Bank in March 2023. While that event was considered shockingly fast, it still required humans to read news and manually interact with banking apps.

In the new scenario described by Barr, there is no need for a human to feel panic or tap a screen. The software simply executes the instructions it was given. This transition from human-led flight to machine-led execution represents a fundamental shift in how financial stress might manifest in the economy.

Synchronized reactions and flaws

The primary concern for regulators is not a single clever AI agent acting alone. The risk lies in thousands of agents running on similar models and being trained on similar data sets. Because these systems often read the same rate sheets and headlines, they are likely to produce synchronized reactions to market news.

When many different traders and customers rely on comparable AI systems, a single flaw shared across those models could have systemic consequences. If a specific signal triggers a sell-off or a deposit move across a wide range of agents, it could create a massive, unified movement of capital. This interconnectedness makes the financial system more vulnerable to uniform software errors.

Furthermore, Barr warned that bad actors could intentionally exploit these synchronized behaviors. By feeding false information into the market, an attacker could mislead many AI systems simultaneously. This would allow them to trigger a mass movement of funds without needing to convince a human crowd to take action.

Current risks and bank defenses

This issue is not a distant theoretical problem for the year 2030; it is already affecting the banking landscape today. Banks are currently seeing AI agents attempt to take actions in retail customer accounts without verified authenticity or proper consent. This highlights a growing problem regarding how agents are granted permissions to move money.

In response to these risks, some major institutions are already taking defensive measures. JPMorgan CEO Jamie Dimon recently revealed that the bank has blocked outside AI agents from customer accounts. Dimon expressed concern over the dangers of having thousands of rogue agents operating inside a major financial company.

The challenge for the industry is balancing the convenience of automated finance with security. While banks want to offer features like sweeping balances to the highest yield, those same features can become channels for rapid deposit flight. Regulators are now looking at how much funding sits in accounts that are reachable by these autonomous agents.

Proposed safeguards and gaps

Barr suggested several practical measures to address these concerns, including strict authentication and consent checks. He also advocated for technical limits on what an agent is allowed to do. These measures are designed to stop impostor agents from accessing accounts and to ensure that legitimate agents act only within specific boundaries.

However, there is a notable gap in these proposed solutions regarding legitimate agents. Authentication can stop a fake agent, but it does nothing to stop a legitimate agent acting on a user's instructions. If ten thousand different legitimate agents all react to the same rate signal at once, the speed of the move remains a problem for the bank.

Currently, the Fed has not published any new rules or circuit breakers in response to these warnings. Barr's remarks serve as a warning to the industry rather than a formal proposal. For fintechs and stablecoin issuers, this means they must consider how their new products might contribute to rapid deposit flight during periods of market stress.

The role of friction and insurance

While the threat of AI-driven runs is significant, certain existing structures may still limit the extent of a crisis. Deposit insurance, emergency lending facilities, and the inherent friction of moving money between different institutions still apply today. These factors may prevent a total collapse even if the initial movement of money happens at high speed.

Banks will likely need to understand their exposure to agent-accessible funds well before the next major interest rate move. The goal for supervisors is to identify which deposits are

Sources and links

First reported by Startup Fortune. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news