This article presents a thought experiment, not a forecast. It does not predict when or whether such an event will occur. Instead, it examines the potential credit consequences of a large-scale technology failure and considers whether current risk analysis and stress testing adequately capture the resulting exposures.
The central hypothesis is that a future liquidity shock could originate not in a deterioration of credit quality, but in the simultaneous failure of technology used by otherwise independent financial institutions.
The relevant historical parallel is the withdrawal of funding during 2007 and 2008. At that time, financing markets closed by sector and then by borrower. The availability of funding deteriorated before prices fully reflected the underlying risks. A comparable dynamic today could originate in a different part of the financial system: the technology used by banks, custodians, payment agents and asset managers to process transactions, maintain records and verify ownership.
A failure in a widely used software component could interrupt payments, compromise the reliability of financial records and create immediate liquidity demands across several solvent institutions. The initial event would be technological, but its consequences could quickly become a credit problem.
The concern is not that artificial intelligence will act independently of human control on a global scale. It is that numerous outsourcing and automation decisions, each reasonable in isolation, may have created common dependencies across institutions that otherwise appear diversified. If those dependencies were disrupted simultaneously, the resulting risks could become correlated.
How the risk could develop
Financial institutions have increasingly outsourced functions such as cloud hosting, identity verification, payment routing, reconciliation, fraud screening and endpoint management. The rationale is straightforward: lower costs, faster software development and, in many cases, more resilient infrastructure.
The collective result, however, may be a concentration of operational risk. Institutions that appear independent can rely on the same technology provider, software library, artificial intelligence model or system configuration. A defective software update, compromised credential repository or unreliable data feed could disrupt multiple institutions at once.
The duration of the disruption would be less important than the extent of the shared exposure. Even a relatively short interruption could create significant liquidity demands if several institutions were unable to establish ownership, reconcile transactions or process payments at the same time.
Automated execution could intensify the disruption. Systems capable of transferring funds, modifying code, accessing production environments or changing user permissions may continue operating while an incident is being investigated. The distinction between systems that make recommendations and those authorized to execute actions is therefore important.
Under normal conditions, automation improves speed and consistency. If the underlying instructions, data or identity controls were compromised, however, automated processes could propagate an error before operators identified the source and intervened. The speed that ordinarily makes these systems valuable could then accelerate the disruption.
What financial authorities have identified
Financial authorities have identified operational resilience, cyber disruption and reliance on external technology providers as areas of concern. The potential systemic risk arises when a disruption affects multiple institutions through the same technology dependencies rather than remaining confined to an individual firm.
These concerns do not require a scenario in which machines take control of financial institutions. They relate to a more conventional problem: failures in operational infrastructure that create funding pressures.
Analysis of financial structures also offers a useful perspective. Off-balance-sheet vehicles, outsourced servicing and leased infrastructure can relocate financial exposures without eliminating them. Some elements of the current investment cycle in artificial intelligence could have similar characteristics. Joint ventures, special-purpose vehicles and vendor-operated infrastructure may distribute obligations across several entities, making the resulting exposures difficult to assess in isolation.
The potential vulnerability lies in the interaction between these structures and a common operational disruption. Obligations that appear manageable under normal conditions could generate simultaneous cash requirements if several counterparties or service providers became unable to perform.
From operational disruption to credit deterioration
The transmission mechanism would probably begin with funding pressure rather than immediate insolvency.
If payment and settlement systems were interrupted, contractual obligations would remain due even when the underlying transactions could not be verified or completed. Institutions might need to post additional collateral against positions they could not reconcile. Customers could withdraw deposits or assets while correspondent banks restricted activity pending confirmation of balances and instructions.
Demand for high-quality liquid assets could consequently rise sharply before any credit downgrade or confirmed financial loss. Market prices would then reflect the uncertainty.
Short-dated bank funding could become more expensive. Credit default swap spreads could widen before losses were established. Additional Tier 1 bank securities could weaken as investors reassessed the potential for coupon suspension, maturity extension or loss absorption, even where reported capital ratios remained above regulatory thresholds.
Three areas deserve particular attention.
First, funding desks that depend on same-day settlement and the rapid transfer of collateral could face immediate liquidity pressure. Second, custodians and payment agents could become critical points of failure where their records are required to establish ownership within a narrow time window. Third, issuers that outsource identity verification and screening could face disruption precisely when reliable checks are most important.
Concentration in enterprise software could also create correlated credit risk among investment-grade companies. A single vendor incident might force several issuers to reduce earnings guidance in the same week, not because demand had weakened, but because billing, provisioning or authentication systems had been interrupted.
Banks could face additional fraud losses and higher verification costs. If voice and remote verification systems were compromised during an increase in artificial intelligence-enabled fraud, operating expenses could rise while customers became less willing or able to transact. The resulting pressure on earnings could precede any deterioration in regulatory capital.
Market conditions could amplify the shock
The effect on credit markets would depend partly on prevailing valuations and liquidity conditions. Neither current credit spreads nor the default-rate forecast indicates an imminent crisis.
Under stress, risk models that appear independent in normal markets may respond to the same signals. Institutions could reduce market-making activity, issue collateral calls and sell assets to raise cash. The resulting demand for liquidity could affect government bonds and corporate credit at the same time, even though investors ordinarily regard these markets as serving different portfolio functions.
Such a move would represent an amplification of the initial disruption rather than an entirely separate financial shock. The concern is that a common operational failure could trigger defensive actions that reinforce one another.
The Anthropic Mythos episode
The April 2026 episode involving Anthropic’s Claude Mythos Preview provides a relevant example of the attention that advanced artificial intelligence capabilities have received from financial policymakers.
In April 2026, United States Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened an emergency meeting with the chief executives of Bank of America, Citigroup, Goldman Sachs, Morgan Stanley and Wells Fargo. The meeting followed the release of Mythos Preview, which Anthropic described as particularly capable of identifying software vulnerabilities.
Anthropic restricted access to a small group of partners, including Amazon, Apple and Microsoft, to give them an opportunity to prepare defensive measures. The episode attracted attention from US financial policymakers concerned about the potential use of advanced artificial intelligence to exploit weaknesses in financial infrastructure.
The episode illustrates the operational concerns that can arise when powerful capabilities interact with widely shared technology dependencies. It also shows why coordination between public authorities and financial institutions may be necessary before a vulnerability results in actual losses.
The episode does not establish that a systemic financial disruption is imminent. It does, however, provide a concrete example of policymakers treating the interaction between advanced technology and financial infrastructure as a potential operational risk.
Lessons from previous banking and market crises
The past two decades provide several useful comparisons. Each illustrates one part of the scenario described here, rather than establishing that the same sequence of events will recur.
2007 to 2008: Funding availability deteriorated before credit losses were fully recognised.
Asset-backed commercial paper programmes and structured investment vehicles that had previously relied on regular refinancing were unable to roll over their funding. Interbank term lending contracted as lenders reassessed counterparties and markets. Institutions that might otherwise have remained solvent faced liquidity difficulties when common exposures to United States mortgage assets were repriced and collateral could no longer be financed on acceptable terms.
The relevant parallel is the possibility of simultaneous liquidity stress without simultaneous insolvency. Funding can disappear before spreads fully reflect the scale of the underlying problem.
2011 to 2012: Common exposures linked banks and sovereigns.
During the euro-area sovereign debt crisis, banks holding substantial amounts of their domestic government’s debt faced correlated funding pressure. The relationship between sovereign credit quality and bank balance sheets became a common driver of risk across institutions.
Long-term central bank refinancing operations and, subsequently, the European Central Bank’s announcement of its Outright Monetary Transactions framework helped stabilise conditions.
The parallel is that one common exposure can influence the funding costs of multiple institutions. An effective response is easier when authorities can rely on reliable records and direct support towards identifiable sources of stress.
March 2020: The dash for cash affected assets normally considered liquid.
Investors sold United States Treasury securities, investment-grade credit and money-market instruments simultaneously to raise cash. Dealer intermediation capacity proved insufficient to absorb the one-way flow.
Official facilities that supported dealers and purchased assets helped restore two-way trading.
The lesson is that the demand for cash can override conventional distinctions between asset classes. Liquidity pressure can cause investors to sell assets that they would otherwise retain, without any corresponding deterioration in the credit quality of those assets.
September to October 2022: Collateral calls created a self-reinforcing cycle in United Kingdom government bonds.
Liability-driven investment strategies came under pressure as long-dated gilt yields rose sharply. Collateral calls forced funds to sell gilts, putting further upward pressure on yields and generating additional calls. The Bank of England temporarily purchased government bonds to interrupt the cycle.
This episode offers a particularly relevant comparison because it combined leveraged exposures, collateral demands and forced selling. The feedback loop was contained through intervention by an external buyer.
The records needed to establish positions and collateral remained available throughout the episode. In the technology scenario considered here, the reliability of those records could itself be in question, potentially complicating the response.
March 2023: Digital banking accelerated the movement of deposits.
Silicon Valley Bank, Signature Bank and First Republic experienced deposit outflows at a pace facilitated by mobile banking and rapid communication. Credit Suisse also faced acute liquidity pressure as confidence deteriorated, despite reported capital ratios remaining above minimum requirements. Its Additional Tier 1 securities were written down as part of the arranged takeover.
The comparison has two elements. Deposits can move rapidly when customers lose confidence, and subordinated bank capital can reprice or absorb losses before a conventional solvency threshold is breached.
Taken together, these episodes show how common exposures can interact with funding mechanisms that assume more time than is available. Official intervention has generally been more effective when authorities could identify exposures, assess reliable records and direct support at a visible source of market stress.
The scenario considered here adds a further complication: the common exposure would be operational, automated processes could continue during the initial disruption, and the records needed to assess positions might temporarily be unreliable.
What regulation already covers, and what may remain untested
Regulatory frameworks address several components of this scenario, although the combined effects of common dependencies, continued automated execution and simultaneous liquidity demands deserve further examination.
In Europe, the Digital Operational Resilience Act has applied since January 2025. It requires financial entities to identify information technology dependencies, assess the potential effects of disruption and demonstrate that important services can remain within defined tolerances. In 2024, the European Central Bank conducted a cyber-resilience stress test focused on firms’ response and recovery capabilities.
In the United Kingdom, the Bank of England, Prudential Regulation Authority and Financial Conduct Authority operate an operational resilience framework centred on important business services and their impact tolerances. A regime for designating critical third-party service providers has supplemented these arrangements since 2025.
In the United States, federal banking agencies updated their third-party risk management guidance in 2023, addressing concentrated reliance on external providers as a safety and soundness concern. The International Monetary Fund and the Financial Stability Board have also identified third-party concentration and cyber disruption as potential channels of systemic risk.
These frameworks have strengthened the way financial institutions identify and manage operational dependencies. The potential gap lies in how risks interact across the system.
Many resilience exercises focus on whether an individual institution can maintain critical services within its tolerance limits. Fewer appear to test a situation in which several institutions activate the same contingency procedures simultaneously, collateral cannot be verified, and automated systems continue to operate before the source of the disruption has been isolated.
The distinction is important for credit analysis. A system may be resilient at the level of an individual institution but less resilient when several institutions depend on the same provider and respond to the same incident in similar ways.
What would weaken the hypothesis?
Evidence that operational controls are keeping pace with technological capability would weaken the scenario.
Relevant indicators would include sustained reductions in software patching and recovery times, supported by independent verification; genuine diversification of critical systems, including tested manual alternatives; and limits on the ability of automated agents to execute sensitive actions without human authorization. Records should also be independently reconstructable when a primary system is unavailable.
Two consecutive quarters of such evidence would provide a useful starting point for assessing whether resilience is improving, although the period alone would not establish that the risks had been eliminated.
Until the evidence is stronger, credit investors could incorporate operational concentration into spread and maturity decisions. Issuer selection should also consider seniority and the extent to which recovery depends on reliable records, functioning settlement systems and the continued availability of third-party services.
Three indicators would suggest that the risk is increasing rather than receding: expanding autonomous execution without equivalent containment controls, growing common dependencies across institutions and technology capabilities advancing faster than patching, oversight and recovery arrangements.
For credit investors, the practical question is straightforward: which issuers could withstand three days without functioning payment and settlement systems?
The answer would depend not only on cash reserves and committed liquidity facilities, but also on the issuer’s ability to establish its positions, access its collateral, maintain essential operations and meet contractual obligations while the disruption is resolved.
Alessandro Rovelli
Rovelli Research
Independent Credit investment research

A major technology outage can travel through contracts and balance sheets long after the system is restored. It's a useful stress-testing question.