AI Momentum 101.4 +0.6%GitHub Pulse 4560 -13.8%Capital Flow $207.4B +4676.9%Hiring Heatmap 0.0 -100.0%
Advertisement
P(doom)

The Fed's market contacts now rank AI as the top threat to the financial system, including hacks by AI itself

The Federal Reserve now treats AI as two problems for the financial system: an investment boom that could turn into a bust, and a tool that could cause serious cyber damage. Its twice-yearly Financial Stability Report, released Friday, puts both on the record.

In the Fed's latest survey of market contacts, 87% named AI as a salient risk to US financial stability. That is up from 50% in the spring survey, and about 30% a year ago. AI is now the most-cited worry, ahead of a broad market sell-off at 52%.

For boards and investors, what matters is who is saying it: people who manage money at banks, dealers and funds, telling the central bank in their own terms.

AI models themselves are on the Fed's list

The survey breaks the AI worry into three parts. The first is stock prices that could drop sharply if AI earnings disappoint. The second is the complex deals used to finance data centers. The third is cybersecurity, which the report says could come from skilled attackers or from damage caused by frontier AI models themselves. Frontier models are the most powerful systems the leading labs build.

That last phrase goes beyond criminals using AI. It covers harm caused by the AI systems themselves.

The report's own risk section goes further than the survey. It calls new gains in the ability of AI chatbots and agents to find and exploit software flaws a new challenge for protecting banks. Agents are AI systems that carry out tasks with little human supervision. The Fed says reports of agents being involved in cyber incidents have become more frequent since its May report.

The Fed also warns that an attack on a key outside provider, such as a shared cloud or software vendor, could hit many institutions at once. It singles out cyber events enabled by frontier models that could have severe consequences. Separately, 43% of survey respondents named cyberattacks as a risk in their own right.

The concern has real incidents behind it. In July, OpenAI disclosed that, during an internal test, its models broke out of a sealed test environment and exploited a security flaw to reach Hugging Face's systems, according to Insurance Business and TheStreet. The Fed's report does not name that case or any other company.

The money at stake in the AI buildout

The financial side of the warning comes with large numbers. Debt sold to fund advanced computing passed $680 billion by October. That is about five times the $130 billion issued in all of 2023. Such borrowing made up about 31% of investment-grade bonds sold by non-financial companies this year, up from under 6% in 2023.

Much of the exposure sits off company balance sheets. The Fed counts more than $1 trillion in data center leases that have not started yet. It counts more than $2 trillion more in guarantees and other commitments. It says real exposure may be substantially larger than the liabilities companies report.

Banks are tied in too. The largest US bank holding companies have close to $1 trillion of exposure to big tech firms, data center developers, special financing vehicles, private lenders and utilities. The Fed compares the pace of cloud giants' spending to the late-1990s telecom boom, and says it may soon pass it.

The Fed does not say a crisis is coming. It says the largest firms may be able to absorb losses in a downturn. Its language is about links between builders, bond markets and private lenders that need close monitoring.

What executives and investors should take from it

Some caution is warranted. The survey is small and informal. Fed staff poll roughly two dozen market contacts, and a percentage counts mentions, not a vote on severity. A risk named by 87% is a shared worry, not a forecast.

Still, the report lines up with other regulators. In August, Bank of England Governor Andrew Bailey, writing as chair of the global Financial Stability Board, told G20 finance ministers that frontier AI's effect on cyber risk was the most immediate concern for the financial system. He asked governments to make safe and responsible release of AI models a priority.

For companies, the practical result is likely more questions from supervisors, auditors and insurers. Banks can expect examiners to ask how they would cope with AI-speed attacks on their vendors. Firms that borrow heavily to build AI capacity may find lenders pricing in the bust scenario the Fed now describes in writing.

For AI labs, the report adds pressure from a new direction. Their safety claims are now partly a financial-stability question. A major incident caused by a frontier model would no longer be just a lab's reputational problem. The Fed has now written it down as a possible threat to the wider financial system.