|
Matthias Knab, Opalesque for New Managers: In the space of four months, autonomous AI agents escaped a test environment and broke into a major technology platform, a state-sponsored cyberattack spoiled food in roughly 700 French supermarkets, a global Microsoft outage knocked out email and collaboration tools in dozens of countries, and every departing flight in the United States was grounded. None of these events moved markets much. According to a new paper by Capital Market Risk Advisors (CMRA), each of them should be read as an early warning.
The paper, titled "New Systemic Market Risks," argues that the threat landscape for investors and financial institutions is shifting. Cyber risk is becoming more aggressive, AI is lowering the cost and skill required to attack, and the same handful of technology companies now dominate both the equity indices and the infrastructure on which markets run. The result, CMRA says, is a form of concentration risk that traditional diversification does not address - and that most stress tests do not capture.
The near-term threat is people using AI, not rogue AI
Could malicious actors use technology and AI to seriously disrupt the economy? CMRA's answer is an unambiguous yes. But the firm is careful to separate science fiction from the more plausible near-term danger. The bigger concern is not AI acting on its own, but malicious people - possibly at state level - using AI to make attacks easier, cheaper and more effective.
That changes the focus of cyber risk prevention. Ransomware and other company-specific threats remain, but institutions now also need to think about system-wide infrastructure disruptions. The central questions, the paper says, are how to anticipate these risks, how to defend against them and how to limit the damage if an attack succeeds.
Risk management is an art as well as a science
CMRA describes risk management as having two dimensions. The first is an art: the imagination to identify unlikely but plausible threats before they materialize, and the judgment to prevent them or reduce their impact. The second is the discipline of measurement and monitoring.
Most major financial institutions now do the second part diligently, producing and reviewing a steady flow of risk dashboards. The danger, according to CMRA, is relying on monitoring while neglecting the imaginative side. "The time to examine vulnerabilities and put effective strategies in place is now," the paper states.
AI is creating a new source of market concentration
The paper places today's market in the early stages of a major technological revolution. Equity returns have rarely been so heavily driven by a small group of large technology companies while much of the rest of the S&P 500 has lagged.
Concentrated leadership is not new in itself. What is unusual, CMRA notes, is that the companies driving returns are taking highly correlated risks on the same technology. That shared exposure is a risk in its own right: what rises together may fall together, and if these few companies decline sharply at the same time, they could drag the broader market down with them.
History shows how fast this can happen. In a crisis, correlations often go to one. Assets that looked unrelated in calm markets suddenly move together as investors sell whatever they can, leverage is unwound and liquidity disappears. It happened with Long-Term Capital Management in 1998, after the failure of Lehman Brothers in 2008, and again in March 2020. Diversification, the paper observes, works best when it is needed least and can fail precisely when it is needed most. A portfolio that looks well diversified based on historical correlations may behave like a single concentrated bet in a crisis.
CMRA is explicit that this is not, by itself, an argument for pessimism or for cutting equity exposure. Markets price expected returns in relation to expected risk, and higher risk should in principle come with higher expected returns. Investors unwilling to bear historically elevated risk can reduce their equity allocations - but they should recognize that doing so may also mean giving up the higher expected returns that come with that risk.
The concentration beneath the market
The less appreciated point, and arguably the core insight of the paper, is that the same concentration exists in the infrastructure underneath the market. A handful of cloud providers, card networks and telecommunications carriers now carry much of the world's economic activity. Several of the companies that dominate the index are also among the largest providers of that infrastructure.
A serious operational failure at one of them could therefore hit both the market's plumbing and its largest holdings at the same time. Diversification across hundreds of stocks offers little protection against a risk that sits underneath all of them. In CMRA's words, correlations can go to one operationally as well as financially: institutions that rely on the same provider can fail together, even if they have nothing else in common.
Three sources of new technology risk
The paper identifies three ways in which technology and AI-related disruptions are likely to arise.
1. Autonomous AI systems behaving in ways their developers did not intend. Long a staple of dystopian fiction, this is no longer purely hypothetical. In July, Hugging Face disclosed that an autonomous AI agent system had breached part of its production infrastructure. OpenAI later confirmed that the agents were its own models which, while undergoing an internal cybersecurity evaluation, had escaped their test environment and chained together vulnerabilities to reach Hugging Face's systems. The controls meant to contain the models were the point of failure. Risk managers will recognize the pattern, CMRA writes: the safeguard everyone relied on was the one that had never been tested under real conditions.
2. Accidental failure of increasingly complex systems. CMRA expects most such incidents to be contained - severe for the organizations directly affected, but limited in their wider economic impact. The nationwide ground stop of U.S. flights on Friday, November 13, may be an example. Whether AI played any role is not known; the FAA attributed the event to a telecommunications failure, and officials said they had seen no indication of a cyberattack. The day before, a global Microsoft outage disrupted Outlook, Teams and Copilot for more than a day, with Microsoft saying there was no sign of malicious activity.
Any single failure may have an innocent explanation. But when multiple critical services fail in quick succession, CMRA argues, the pattern deserves scrutiny rather than dismissal as coincidence. At a minimum, it shows how much now depends on a few providers.
3. Deliberate attack. This is perhaps the most serious near-term source of risk. France has formally blamed Iran's Revolutionary Guard for an attack on the night of October 11-12, in which hackers reportedly took control of refrigeration systems at about 700 supermarkets and other retail locations and raised temperatures in coolers and freezers. Food spoiled, losses reached tens of millions of euros and at least 15 fires broke out. The controllers were accessible over the internet, which allowed the attackers to reach and change their settings.
Nothing reported so far indicates that AI was used in the French attack. But AI can lower the skill, time and cost needed to find and exploit vulnerabilities of exactly this kind. The paper asks readers to consider the implications if a hostile state, a terrorist group or even a disgruntled individual could use AI to disrupt a country, an institution or an economy - or simply to cause damage.
Which systems would cause the most disruption?
Some systems are so embedded in economic activity that an attack or technical failure would have consequences far beyond the organization directly hit. CMRA singles out four categories:
Electronic payments. A disruption of the major credit and debit card networks would severely affect commerce, and cash and cheques could offer only a limited short-term substitute at today's scale. A broader failure of electronic payments - including bank transfers and the other channels through which businesses and financial institutions move money - could bring large parts of commerce to a standstill, even if the banks themselves remained financially sound.
Market data, pricing and collateral systems. Valuation, margining and risk reporting depend on a small number of data vendors, pricing services and clearing systems. If they fail, institutions may be unable to mark positions, calculate margin or know their exposures - precisely when markets are most volatile.
The internet and its supporting infrastructure. Rather than imagining the entire internet being shut down, CMRA suggests considering a major disruption to critical internet infrastructure, telecommunications providers or the large cloud platforms. Such an event could hit thousands of otherwise unrelated organizations at once, and the consequences would be amplified by how many other essential services depend on that infrastructure.
Institution-specific systems. Consider a bank whose capital, liquidity and customer assets are intact, but whose computer systems have been compromised or are unavailable. It may be unable to access customer accounts, process payments, execute transactions or communicate with counterparties. The institution would remain financially sound yet be unable to do business. Identifying such vulnerabilities, assessing the consequences of their failure and strengthening defenses before they are tested is, in CMRA's view, the risk manager's core task.
When the systems stop, the markets don't
Drawing on decades of experience in derivatives disputes, CMRA makes a point that is often overlooked: markets keep moving while institutions are stuck. In the Canadian asset-backed commercial paper crisis, a 17-month standstill that began in 2007 shaped the outcome of later disputes - the parties stood still, but the markets did not. (CMRA assisted in the allocation of damages in Barclays vs. Metcalfe & Mansfield, the Devonshire Trust case.)
An institution that cannot access its systems faces the same problem in compressed form. It may be unable to trade, value positions, meet or make margin calls, or even receive notices, while its exposures continue to change.
And when the systems come back, the lawyers arrive. CMRA, which has acted as an expert witness since the 1990s in matters ranging from Orange County and LTCM to engagements arising from Lehman, "Volmageddon" (the February 2018 volatility shock) and COVID, expects outages of this kind to raise familiar questions:
- Default or force majeure? Was a missed payment or delivery an event of default, or an event outside the party's control? The answer may depend on which version of the ISDA Master Agreement governs and what the parties negotiated.
- Valuation when prices disappear. How should positions be valued for close-out or collateral purposes when markets are dislocated, data feeds are down or prices cannot be observed?
- Commercial reasonableness. Did the parties act reasonably given what they knew, and could know, at the time?
- Notice and communication. Can a valid notice be delivered when email, messaging platforms or the counterparty's systems are unavailable?
The time to review documentation, valuation policies and fallback procedures, the paper stresses, is before the disruption, not after.
Data is king
Courts and arbitrators need to understand what the market looked like at the precise moment a margin call was made, a close-out valuation was taken or a notice was sent. Intraday pricing data from a period of extreme stress is notoriously hard to reconstruct after the fact, and the reconstruction gets harder as time passes. A technology outage makes this worse still, because the systems that would normally capture the record may be the very ones that failed.
CMRA therefore recommends that institutions arrange now for market and transaction data to be captured independently of their primary systems, and that they preserve, during and after any disruption:
- Market data: intraday prices, volatility surfaces, bid/offer spreads, order book depth and open interest, from more than one source where possible.
- Communications: broker and dealer communications, axe sheets, margin calls and notices, with a record of when each was sent and received - or failed to arrive.
- Internal records: risk escalation records, valuation decisions and the reasoning behind them, and contemporaneous notes of who knew what and when.
- The outage itself: system logs, incident timelines and vendor status reports showing which systems were down, for how long and why.
Parties that fail to preserve this record will be at a significant disadvantage when experts later reconstruct what the market was saying at the time. Data preservation, CMRA says, should be treated as a live exercise, not a retrospective one.
Build resilience before the next disruption
No single company can prevent failures affecting infrastructure as large as the payment networks, telecommunications systems or the internet. But every institution should have a credible plan for continuing essential operations if these systems become unavailable. That means identifying critical functions, establishing alternative ways of performing them, and regularly testing the ability to operate under adverse conditions.
Organizations can also reduce their exposure. Systems that do not need a constant external connection should be isolated from the internet, or designed so they can be disconnected quickly. Firewalls, network segmentation and, where appropriate, air-gapped systems help protect critical infrastructure from sophisticated attacks. "Do we need fax machines again?" the paper asks, only half in jest.
CMRA also recommends a layered architecture: an outward-facing system handles communication with the outside world, while access to critical internal systems is restricted to a small number of closely monitored connections. Critical data should be backed up in secure, isolated environments so operations can be restored even if primary systems are compromised. Together with regularly tested recovery procedures, these are the essential elements of defensive systems engineering.
Major financial institutions, software companies, AI developers and governments should all begin strengthening their defenses against the malicious use of AI now. For investors, the lesson is that market risk in 2026 and 2027 may come not only from valuations, leverage or the economic cycle, but also from failures or attacks on the technological infrastructure on which financial markets and the wider economy depend.
Stress test your assumptions, not just your positions
Most stress tests shock positions: rates up, equities down, spreads wider. Far fewer stress the assumptions on which the entire risk framework rests. Yet the largest losses rarely come from a risk nobody measured. They come from an assumption everybody took for granted - that hedges would work, that liquidity would be there, that a counterparty would survive, that volatility would stay within its historical range. The collapse of short-volatility strategies in February 2018 showed how quickly positions built on calm-market assumptions can lose most of their value.
CMRA urges every institution to list the assumptions behind its risk models, valuation policies and contingency plans, and to ask what happens if each one fails, alone and in combination. Reverse stress testing is a useful discipline here: start from an outcome the institution could not survive and work backward to the combinations of events that would produce it. Historical data is a starting point, not a ceiling. The next crisis will not be a replay of the last one.
The paper pairs common assumptions with the stress question each should prompt:
"Diversification will protect us" - What if every position falls together and correlations go to one?
"Liquidity will be there" - What if we must sell, or post collateral, when no one is buying?
"Our hedges will work" - What if the hedge and the position move the same way, or the hedge counterparty fails?
"Counterparties will perform" - What if a major counterparty cannot pay, or cannot even operate its systems?
"Prices will be observable" - What if pricing feeds go down or markets gap, and we cannot value positions or margin?
"Our systems and vendors will be up" - What if our cloud, telecom or data provider is out for 72 hours during a market shock?
"We can always communicate" - What if email and messaging fail just as notices and margin calls must be sent?
Six questions boards should be asking now
The paper closes with a checklist for boards and senior management:
1. Which of our critical functions depend on a single cloud, telecommunications or payments provider?
2. If our systems were unavailable for 72 hours, how would we value positions, meet margin calls and communicate with counterparties and regulators?
3. Do our ISDA and other trading agreements address force majeure, notice and valuation in a prolonged outage, and do we know which version governs each relationship?
4. Where are we using AI agents, what can they access, and who has tested whether they can be contained?
5. If we had a dispute arising from an outage, could we prove what the market looked like and what we did, minute by minute?
6. When did we last run a scenario that combined a market shock with an operational outage?
The big picture
The common thread running through CMRA's paper is that the next systemic event may not look like the last one. The concentration that investors worry about in the index has a mirror image in the cloud, payment and telecom infrastructure beneath it - and several of the same companies sit on both sides. An institution can be solvent and still be unable to operate; a portfolio can be diversified and still behave like a single position. The incidents of the past four months did little damage to markets. Their value lies in what they reveal about where the next shock could come from, and how little time institutions may have to prepare once it arrives.
About CMRA
Capital Market Risk Advisors (CMRA) has provided risk advisory and expert witness services since 1992. The firm helps boards, investors, financial institutions and their counsel identify, measure and manage risk across what it calls its "Galaxy of Risks," and has served as expert witness in many of the landmark financial disputes of the past three decades. More information: www.cmra.com
|