Read the full analysis: How The Fed’s 87% AI Risk Finding Connects To Australia’s Stress Test Plan on ThorstenMeyerAI.com
TL;DR
The Federal Reserve’s November 2026 Financial Stability Report says 87% of 23 market contacts cited AI as a risk to U.S. financial stability, up from 50% in the spring. Australia’s APRA says frontier AI models are accelerating cyber threats and plans to examine AI-driven disruption in its 2027 system-wide stress test.
The Federal Reserve’s November 2026 Financial Stability Report says 87% of 23 market contacts cited artificial intelligence as a risk to U.S. financial stability, while Australia’s prudential regulator APRA says its 2027 system-wide stress test will examine AI-driven disruption. The developments point to regulators’ concerns about both the financial exposure behind the AI buildout and the possibility that AI-enabled cyber incidents or a failure at a shared provider could affect multiple institutions.
The Federal Reserve Bank of New York gathered responses from market contacts during September and October. In the survey, 87% named AI, up from 50% of 20 contacts in the spring. The panel included professionals at broker-dealers, banks, investment funds and advisory firms. The Fed cautions that the survey summarizes respondents’ views and does not represent the views of the Federal Reserve. Its report reflects information available through October 28, before events in November.
Respondents described several distinct AI-related risks: a possible correction in AI-linked equities if large investments fail to produce expected returns; cyberattacks made easier by AI, including potential unintended damage from frontier models; and complex financing tied to AI. Other leading risks included a correction in risk assets, cited by 52%; higher long-term interest rates, 48%; cyberattacks, 43%; and fiscal debt sustainability, 43%.
The report also quantifies borrowing connected to advanced computing. Issuance passed $680 billion by October 2026, compared with $130 billion over all of 2023. It represented about 31% of investment-grade nonfinancial corporate issuance, up from less than 6% in 2023. Separately, the report says the largest U.S. bank holding companies have close to $1 trillion in exposure to firms and vehicles involved in the buildout.
Explained in motion
How one shared AI provider can ripple through finance
Banks, insurers and funds often rely on the same few models, platforms and service providers. Supervisors warn that one problem there can reach all of them at once.
Illustrative sequence, not a real incident
Shared AI provider
Bank
Insurer
Pension fund
Payments
Customers and markets
Shared providerMany institutions run on the same model, platform or service provider.
One changeA failure, a compromise or a model update changes how that provider’s AI works.
Many hit at onceEvery institution that depends on it is affected at the same time.
Services stallPayments, claims or trading slow down. The Fed warns that disruptions can hurt market liquidity and confidence.
No quick switchThe damage grows when there is no ready substitute for the service.
Source: APRA, System Risk Outlook (November 12, 2026) and letter to industry on AI (April 30, 2026); Federal Reserve, Financial Stability Report (November 2026), pp. 47–48. Illustrative sequence built from the mechanisms they describe.
Graphic: Powered by Thorsten Meyer AI
Why Shared AI Risks Matter
The concern is not limited to whether AI-related companies’ valuations hold up. The Fed describes a potential chain in which a cyber incident at a critical third-party service provider affects several financial institutions at once, especially if those institutions cannot readily substitute another provider. The report says disruption could weaken market liquidity, erode investor and depositor confidence, and prompt forced asset sales.
That makes concentration in technology suppliers a financial stability issue as well as an operational one. APRA’s outlook adds a practical test of that exposure: it says few institutions had tested whether they could switch away from critical AI providers. If those arrangements fail during a disruption, firms and customers could face service interruptions beyond the organization where an incident began. The regulators’ statements identify risks, however; they do not establish that a specific AI provider has caused a financial system disruption.
Two Regulators, Different Evidence
The Fed’s 87% figure comes from a small survey panel, not a measurement of the probability that AI will cause a crisis. The spring comparison was based on 20 contacts surveyed from March through April; the latest round had 23 respondents. The survey records professional judgments about possible shocks over the next 12 to 18 months. Its results should be read as a change in the concerns of those respondents, not as a forecast or a representative poll of the entire financial sector.
The Fed’s report separately says the U.S. banking system is “sound and resilient.” It notes that equity valuations remain high, partly amid market enthusiasm for superintelligence, and warns that advances in large language models and agentic systems have introduced new challenges in protecting financial institutions, infrastructure and third-party providers. It says reports of agentic systems being involved in cyber events have increased since the May report, but gives no count.
APRA’s November 12 System Risk Outlook focuses on Australia’s system-wide preparedness. It names frontier models from Anthropic and OpenAI as examples of systems that, in APRA’s assessment, are improving at finding and exploiting software vulnerabilities. APRA’s planned 2027 exercise will examine AI-driven economic disruption. That plan is not the same evidence as the Fed’s survey: one is a regulator’s assessment and announced test scope, while the other is a poll of market contacts.
“Frontier AI models, including those from Anthropic and OpenAI, are speeding up cyber threats.”
— APRA System Risk Outlook, November 2026
Limits of the Risk Evidence
The Fed does not identify its 23 respondents, and the survey does not show how likely any of the listed risks are to occur. Its 87% figure is a share of participating contacts, not the share of financial firms, assets or the public exposed to AI risk. The report also does not provide a count for the increased reports of agentic systems’ involvement in cyber events.
APRA’s announced test scope does not yet establish which scenarios will be included, how institutions will be assessed or what findings will be published. The source material also does not specify how either regulator will measure dependence on common AI providers or the ability to switch services. The cited concerns are supervisory assessments and respondent views, not confirmation that a particular AI model or provider has caused systemic harm.
What the 2027 Test Could Show
APRA’s next stated milestone is its 2027 System Stress Test, which is to examine AI-driven economic disruption. Details of the scenarios and timetable are not provided in the available material. The test may give regulators and participating institutions a structured way to examine how AI-related disruption could affect operations and the broader system, but its conclusions are not yet known.
In the United States, the November report presents the latest survey and risk assessment; the source material does not announce a specific follow-up exercise or policy change. Readers should distinguish future supervisory work from current findings: the Fed has documented contacts’ concerns and financial exposures, while APRA has announced a test plan. How either regulator’s next steps address shared-provider dependence remains to be seen.
Key Questions
What does the Fed’s 87% figure measure?
It is the share of 23 market contacts surveyed in September and October who named AI as a risk to U.S. financial stability. It is not a measure of the probability of a crisis or a survey of all financial institutions.
What AI-related risks did respondents identify?
They cited possible corrections in AI-related equities if returns disappoint, AI-enabled cyberattacks—including possible unintended damage from frontier models—and complex financing tied to AI. The Fed also reported substantial borrowing connected to advanced computing.
What will Australia’s APRA stress test examine?
APRA says its 2027 system-wide stress test will examine AI-driven economic disruption. The available material does not detail the test scenarios or assessment method.
Does the Fed report say AI has caused a financial crisis?
No. It reports concerns expressed by market contacts and describes potential vulnerabilities. The report does not claim that AI has caused a financial crisis or confirm that a specific provider has triggered systemic harm.
Source: ThorstenMeyerAI.com