AI

How to Close AI’s Accountability Loophole

Published

1 month ago

May 28, 2026

On 14 May 2026, legal scholars gathered in New Delhi for the International AI Accountability Forum with a question that every major economy has, until recently, chosen to defer. An autonomous AI agent had concluded a commercial contract on behalf of a firm without any human reviewing the terms. The deal violated an obscure antitrust provision. No one was certain who bore responsibility — the developer who built the model, the enterprise that deployed it, or the executive who had simply clicked “enable autonomous mode” one Tuesday morning and moved on to something else.

That ambiguity is no longer an edge case. It’s the operating architecture of global commerce in 2026.

The Governance Gap That Grew While Nobody Was Watching

For three years, the dominant narrative in AI policy was one of cautious progress. Frameworks were published. Principles were endorsed. Voluntary codes of practice were signed — or, in the case of Meta, pointedly declined. The EU AI Act entered into force in August 2024, its obligations phasing in through 2027 in a risk-tiered structure that many compliance teams privately described as sensible. American legislators, meanwhile, produced a patchwork of state laws — Colorado’s AI Act, California’s AB 2013, Texas’s Responsible Artificial Intelligence Governance Act — that created meaningful but geographically fragmented protections.

The problem is that the technology didn’t wait for the law to catch up.

Non-human and agentic AI identities are projected to exceed 45 billion by the end of 2026 — more than twelve times the entire human global workforce. Enterprises are now contending with an 82:1 ratio of autonomous AI agents to human employees, according to Palo Alto Networks. Yet only 44% of organisations have formal AI governance policies in place. That 38-percentage-point chasm is not a statistic. It’s a liability map.

The Anatomy of the AI Accountability Loophole

The AI accountability loophole does not arise from malice. It arises from architecture. Earlier generations of AI advised humans, who then acted. Contemporary agentic systems receive a goal, decompose it into sub-tasks, execute against real-world environments — APIs, financial platforms, hiring databases, supply chains — and adapt their behaviour in response to outcomes. The original human instruction becomes increasingly remote from the final, potentially harmful output.

Legal scholars call the resulting liability void a “moral crumple zone”: responsibility diffuses across developers, operators, and deployers, with no single party absorbing it cleanly. Courts, trained on centuries of product liability doctrine in which a manufacturer and a product could be causally linked, are poorly equipped to adjudicate what amounts to an emergent harm from a multi-party autonomous chain.

The agentic AI liability gap is already appearing in commercial practice. Clifford Chance noted in February 2026 that legacy technology agreements — designed for software operating under human direction — say virtually nothing about a customer’s rights to understand or control an AI agent’s behaviour. Yet, when something goes wrong, the deployer must justify that behaviour to regulators, auditors, and courts. The GDPR’s transparency and explainability obligations fall on the enterprise. The contract with the AI vendor may offer none of the audit rights those obligations require.

The January 2026 OpenClaw incident illustrated this with uncomfortable precision. The firm’s AI assistant leaked sensitive credentials across multiple messaging platforms — not because the system malfunctioned, but because it executed its instructions exactly as designed. No one had defined the boundaries. No one had established who would be responsible when autonomous actions spiralled past their intended scope.

This is the structural truth of the loophole: it doesn’t look like a failure until it’s too late to prevent one.

What is the AI accountability loophole, and why does it matter? The AI accountability loophole is the legal and governance gap between deploying autonomous AI systems that take real-world actions and establishing documented frameworks that assign liability when those actions cause harm. It matters because, as of 2026, 82% of organisations use AI agents while only 44% have formal governance policies, leaving the majority operating with live exposure and no clear accountability chain.

Why Existing Regulation Doesn’t Yet Reach the Problem

The EU AI Act is the most serious attempt yet to impose structural accountability on AI — and it’s worth understanding precisely where it reaches and where it falls short.

The Act’s general-purpose AI rules became legally applicable on 2 August 2025. The European Commission’s enforcement powers, however, don’t come into force until 2 August 2026. That year-long gap — obligations without enforcement — created a predictable compliance posture: many providers engaged with the Act’s Code of Practice in good faith, but the absence of live penalty risk reduced urgency. Finland became, in January 2026, the first EU member state with fully operational AI Act enforcement powers at the national level. The rest of the bloc has yet to fully follow.

The Act’s penalties are real enough: up to €35 million or 7% of global turnover for the worst violations. Yet the Act does not yet define “agentic AI” as a distinct category. Existing high-risk classifications apply based on what the agent does, not on how it’s labelled. An autonomous agent executing hiring decisions falls under high-risk AI rules. The same agent executing supply-chain procurement decisions may not. That definitional seam is where sophisticated legal teams will probe for exits.

The US situation is, if anything, less coherent. As of April 2026, no comprehensive federal AI liability law has been enacted. The Trump administration’s March 2026 National Policy Framework for Artificial Intelligence called for a single federal approach with guardrails around child safety, intellectual property, and national security — a framework designed as much to preempt state-level activity as to govern AI itself. Congress is debating next steps, but the divergence between the EU’s precautionary architecture and Washington’s innovation-first instincts is structural, not accidental.

China, for its part, governs AI through targeted rules emphasising social stability and content control. For multinationals, that means three distinct and partially contradictory accountability architectures operating simultaneously — each with different transparency requirements, different liability triggers, and different enforcement bodies.

The picture is more complicated still when insurance enters the calculation. Verisk introduced optional generative AI exclusions effective January 2026, covering 82% of global property-casualty templates. The market is, in effect, pricing in the loophole before the law has closed it.

The Case for Minimal Regulatory Interference

The accountability-first position has a coherent opponent, and it deserves a fair hearing.

A significant constituency in Washington, parts of the UK government, and much of the venture community argues that liability-heavy regulation will simply export AI development to jurisdictions with lighter governance. The Trump administration’s framework explicitly framed AI regulation in national-security terms: the US cannot afford to constrain domestic frontier AI development while China runs an integrated state-industry model with no comparable friction. Meta’s decision to decline the EU’s GPAI Code of Practice — citing concerns about legal uncertainty and scope — reflects a calculation that voluntary compliance costs are real, while the benefits of safe-harbour protection are theoretical until enforcement bodies have track records.

There’s a serious point embedded in the industry position on foreseeability. The standard product-liability doctrine requires that harm be foreseeable by the manufacturer. Autonomous AI systems operating in novel, unscripted environments produce outcomes that are genuinely difficult to anticipate by design — that emergent capacity is what makes them commercially valuable. Holding developers strictly liable for unforeseeable harms from systems their customers then modify and deploy could be not only legally questionable but economically chilling.

Still, the counterargument has force. The EU’s forthcoming Product Liability Directive, effective December 2026, explicitly includes software and AI as “products” under strict liability doctrine. If a system is found defective, the manufacturer’s liability doesn’t depend on the customer’s foreseeability; it depends on whether the system met its safety specification. That framework is workable. What it requires is that developers and deployers actually specify what their systems are supposed to do — a baseline that many current agentic deployments conspicuously lack.

What a Real Fix Looks Like

The conceptual path forward exists. Singapore’s IMDA Model AI Governance Framework for Agentic AI, published in 2025, introduced the concept of Meaningful Human Control — defined as the unity of human understanding, intervention capacity, and traceability of responsibility. It’s a cleaner formulation than anything currently embedded in EU or US regulation. The question is whether it can be translated into enforceable obligation across multiple jurisdictions, rather than remaining one more well-intentioned framework on a shelf of well-intentioned frameworks.

Three operational changes would close the loophole more quickly than any single piece of legislation.

The first is mandatory decision logging. Boards are already beginning to require that every autonomous agent maintain a cryptographically secured record of the inputs, model weights, and logic used to reach a consequential output. Without such a log, neither courts nor regulators can trace harm to a specific decision node. The EU AI Act already mandates logging for high-risk AI systems; extending that mandate to all agentic systems operating above a defined authority threshold would remove the definitional ambiguity.

The second is contractual restructuring. Clifford Chance’s February 2026 guidance put it plainly: enterprises must renegotiate vendor agreements to expand indemnities, lift liability caps, and impose explicit audit rights over AI agent behaviour. That’s not a regulatory requirement — it’s a commercial one, enforceable through the existing law of contract.

The third is the least glamorous and probably the most important: OWASP’s Least-Agency principle. An AI agent should hold the minimum autonomy and access necessary for its defined task, and no more. The OWASP Top 10 for Agentic Applications 2026 — compiled with input from over 100 industry experts — identified Tool Misuse and Identity and Privilege Abuse as the second and third most critical risks in agentic systems. Both trace directly to agents holding more permission than their task scope requires. This is not a regulatory problem. It’s an engineering decision made at the time of deployment.

The Accountability Reckoning Ahead

The August 2026 activation of the European Commission’s full enforcement powers against GPAI model providers marks a genuine inflection. Regulators will be able to request documentation, conduct evaluations, order model recalls, and impose fines. For the first time, the gap between obligation and enforcement will close — at least in Europe, at least for foundation models, at least for now.

That’s a narrower set of “at leasts” than the moment requires.

The deeper problem is that the AI accountability loophole isn’t primarily a European problem or an American one. It’s a product of deployment velocity that has outrun every governance institution on the planet simultaneously. Organisations are embedding autonomous systems into consequential decisions — financial, medical, legal, logistical — faster than any single regulatory body can audit, and faster than most legal teams can document.

The liability exposure exists now. It doesn’t wait for regulatory clarity to materialise. Courts in California have already demonstrated willingness to hold deployers accountable for AI hiring tools that discriminate; the plaintiff’s bar in New York and Brussels has watched those cases closely. The insurance market has moved to exclude the risk. The question for every board with significant AI deployment is not whether accountability frameworks are coming. It’s whether they’ll arrive before or after the claim does.

Autonomous systems that act in the world must be owned by someone who can be held to account in the world. The technology to build such systems has outpaced every institution designed to govern them. That gap is the loophole — and the work of closing it can’t wait for the next summit.

Discover more from The Economy

Subscribe to get the latest posts sent to your email.

Up Next

Russia Overspends on Putin’s War in Ukraine by $28bn

Don't Miss

Pakistan Budget FY 2026-27: Relief, Prospects, and the Tightrope Walk

Click to comment

AI

AI Capex Bubble 2026: The Hidden $662B Debt Nobody Reports

Published

2 hours ago

July 11, 2026

Abdul Rahman

Every earnings season now brings a fresh wave of headlines about hyperscaler AI capital expenditure hitting a new record. The “big four” — Amazon, Microsoft, Alphabet, and Meta — are on track to spend roughly $725 billion combined in 2026, a 77% jump from the $410 billion deployed in 2025 (UnboxFuture). That number gets reported constantly. What almost nobody is reporting with the same prominence is a separate figure that may matter more: roughly $662 billion in data center lease commitments that hyperscalers have already signed but not yet begun — obligations that currently sit entirely off balance sheet.

Why the Off-Balance-Sheet Number Changes the Whole Picture

Under GAAP accounting rules governing when a lease “commences,” these signed-but-not-started commitments don’t appear in the capital expenditure figures analysts and investors typically scrutinize when assessing hyperscaler financial health. According to reporting citing Moody’s early-2026 analysis, this shadow liability is larger than the combined on-balance-sheet debt of the same companies (Anomaly Investments).

That detail matters enormously for one specific argument AI infrastructure bulls have relied on: the claim that this buildout is being conservatively self-funded from operating cash flow rather than risky leverage. Once the full picture of committed-but-unrecognized obligations is accounted for, that defense becomes much harder to sustain.

The Debt Is Already Showing Up, Not Just Theoretical

This isn’t a purely hypothetical concern about future liabilities. Big tech companies have already issued more than $100 billion of bonds in 2026 specifically to help fund AI capital expenditure, and investors have responded by demanding record levels of protection against potential defaults through credit default swaps — essentially insurance policies against bond default (IEEE ComSoc).

Individual company examples illustrate the shift toward leverage: Oracle issued an $18 billion bond specifically tied to its data center expansion; CoreWeave secured a $2.6 billion loan alongside a $1.75 billion bond package; and OpenAI and Oracle reportedly entered into a $100 billion vendor financing arrangement (Anomaly Investments). At Amazon specifically, capital expenditure over the trailing twelve months has reached $151 billion — a figure that now exceeds the company’s entire operating cash flow, pushing free cash flow into negative territory.

The Depreciation Assumption Almost No Coverage Questions

Here’s an angle genuinely underexplored across most financial media: the depreciation schedules hyperscalers use for AI hardware assume a five-to-six-year useful life. But given how rapidly GPU generations are turning over and how intensively AI workloads are pushing hardware utilization, critics argue the real economic life of this equipment is closer to two to three years. That gap between assumed and actual depreciation is estimated to understate true asset depletion by roughly $176 billion between 2026 and 2028 alone — a figure that grows as accelerating token consumption pushes hardware utilization beyond the assumptions built into current depreciation schedules (Anomaly Investments).

Layered on top of that is the energy cost curve: running the current roughly 30-gigawatt installed base of AI infrastructure costs approximately $27 billion annually today, but that figure is projected to climb to between $45 and $90 billion per year as capacity scales toward 2029 — and crucially, these are first charges against revenue, not optional or deferrable costs.

The Revenue Gap: Who’s Actually Paying for All This?

The most commonly cited justification for the capex surge is that the pure-play AI vendors — OpenAI, Anthropic, and others — represent a massive and rapidly growing revenue opportunity. The reality is more nuanced. OpenAI’s roughly $20 billion annualized revenue run rate, while genuinely impressive for a company with barely any consumer products three years ago, represents only about 3% of projected 2026 hyperscaler capex. Anthropic’s roughly $9 billion run rate, despite showing 9x year-over-year growth, occupies a similarly small share. The entire cohort of pure-play AI vendors combined — including Cohere, Mistral, Perplexity, and others — likely accounts for less than $35 billion in projected combined 2026 revenue against a hyperscaler capex figure exceeding $700 billion (Futurum Group).

That gap is the crux of the bubble debate: hyperscalers are betting the infrastructure will ultimately serve enterprise adoption and their own AI services broadly, not just third-party AI vendor revenue — but that bet requires enterprise AI monetization to arrive at a scale that, as of mid-2026, remains largely unproven outside of code generation and basic customer service automation.

The Skeptic’s Case, From Inside Goldman Sachs Itself

The most prominent voice of institutional skepticism doesn’t come from an outside critic — it comes from within Goldman Sachs itself. Jim Covello, the bank’s Head of Global Equity Research, has consistently argued the economics of the generative AI transition are fundamentally flawed, stating in mid-2026 that the industry has moved “further away” from justifying the scale of capital expenditure compared to two years prior (UnboxFuture). Covello has specifically flagged circular capital flows between cloud providers and AI startups — where hyperscalers invest in AI companies that then spend that same capital purchasing compute from those same hyperscalers — as a red flag reminiscent of vendor financing patterns seen in the dot-com era.

The valuation comparison to that era is explicit and increasingly common among strategists: US technology and AI equities carry EV/EBITDA multiples near 25x, close to historical extremes and above the telecom valuations that preceded the 2000 dot-com peak. More specifically, capex is currently expanding roughly 46 percentage points faster than revenue growth — a gap that exceeds the 32-point divergence observed during the 2001 telecom excess cycle (Allianz Research). Separately, Bank of America strategists have pointed out that AI stock concentration has reached levels matching prior bubble peaks, with the “AI Big 10” (Nvidia, Microsoft, Alphabet, Amazon, Meta, Apple, Tesla, Broadcom, Micron, and AMD) now making up 41% of the S&P 500 — comparable to the concentration of tech and telecom stocks during the actual dot-com bubble (Yahoo Finance).

The Bull Case Isn’t Naive Either

It would be inaccurate to frame this purely as informed skeptics versus blind enthusiasm. Goldman Sachs’ own broader research (distinct from Covello’s individual view) models roughly $7.6 trillion in cumulative AI capital expenditure between 2026 and 2031, built on the expectation that token consumption will increase 24-fold by 2030, driven largely by enterprise AI agents becoming embedded in production workflows rather than remaining experimental (Sesame Disk / Goldman commentary). Microsoft has disclosed an $80 billion backlog of Azure orders it currently cannot fulfill due to power constraints — genuine evidence that demand, at least for existing capacity, is outpacing even the current aggressive build-out pace (Futurum Group).

Leverage levels also remain more conservative than headlines suggest in absolute terms: the top five US capex providers reported a combined $385 billion in debt at the end of 2025, with leverage ratios still roughly 20% below the “high spender” cohort from the 2000 dot-com peak, according to Allianz Research analysis — meaning rising debt levels are a trend worth monitoring closely, not yet an acute crisis.

What Happens If the Bubble Skeptics Are Right

Historical infrastructure cycles offer a specific and somewhat counterintuitive lesson: the investors who fund the initial frenzied build-out phase rarely capture the long-term rewards. If the AI capex cycle follows the pattern of the 1998-2001 fiber optic buildout, hyperscalers may eventually be forced to write down the value of data centers and GPUs purchased at today’s prices and utilization assumptions. But that collapse in computing costs, paradoxically, could pave the way for a new generation of leaner, genuinely profitable software companies to build on top of the resulting cheap, overbuilt infrastructure — much as fiber-optic overbuild eventually enabled the 2000s streaming and cloud computing boom, even after the original telecom investors were wiped out.

What This Means for Investors and Businesses

For equity investors, the practical signal to watch isn’t the headline capex number — it’s the widening gap between capex growth and revenue growth, and whether that gap begins narrowing through 2027 as enterprise adoption either accelerates or disappoints. For businesses evaluating AI vendor relationships, the circular-financing pattern flagged by Covello is worth diligence: understanding whether an AI vendor’s revenue depends partly on capital originally supplied by the same hyperscaler providing its compute is a legitimate red flag for assessing that vendor’s underlying financial independence. For fixed-income investors, the rising credit default swap pricing on hyperscaler-linked debt is itself a market signal worth tracking as an early indicator of shifting sentiment, independent of equity price action.

The Bottom Line

The AI infrastructure buildout genuinely is the largest corporate capital expenditure cycle in recorded history, and it’s happening for real, defensible reasons tied to a genuine technology shift. But the debate over whether it constitutes a bubble isn’t really about whether AI technology is useful — it’s about whether the timing of returns can keep pace with public equity markets’ patience, and whether the $662 billion in off-balance-sheet lease commitments, aggressive depreciation assumptions, and circular vendor financing arrangements represent manageable financial engineering or the early architecture of a genuinely serious correction. Both cases have real evidence behind them. What’s clear is that the headline capex figure everyone quotes is no longer the most important number in this story.

Discover more from The Economy

Subscribe to get the latest posts sent to your email.

AI

AI Bubble Warning 2026: Why BIS, IMF and Bank of England Fear a Market Crash

Published

6 days ago

July 5, 2026

Abdul Rahman

Global financial regulators have moved from quiet skepticism to open warning, marking one of the most significant shifts in central-bank rhetoric since the aftermath of the 2008 crisis. The Bank for International Settlements (BIS), the International Monetary Fund (IMF), and the Bank of England have each flagged the risk that a correction in artificial-intelligence valuations could cascade through the global financial system, according to the BIS Annual Economic Report 2026 and reporting compiled by Wikipedia’s tracking of the unfolding episode.

From Confidence to Contagion Fear

The warnings did not emerge in a vacuum. In late June 2026, South Korea’s KOSPI index was forced into a trading halt after Samsung and SK Hynix shares each lost roughly 12% in a single morning, a shock that rippled into the Nasdaq, which fell 2.2% the same day. By the following week, Oracle had recorded its worst trading week since the dot-com crash, sliding 19%, after Apple raised product prices in response to soaring chip costs. The sell-off, detailed in Wikipedia’s account of the June 2026 rout, spread across global chip manufacturers before the BIS issued its formal caution on June 29.

Pablo Hernández de Cos, general manager of the BIS, framed the moment as one of “progress” colliding with “peril,” pointing to inflationary pressure, elevated public debt, and what the institution calls AI exuberance as compounding financial vulnerabilities.

Why This Cycle Looks Different — and Why It Doesn’t

Comparisons to the 1999–2000 dot-com bubble are now routine among Wall Street strategists. Deutsche Bank’s global economics team has described 2026 as resembling “1999 meets 1990,” according to Fortune’s coverage of the growing exuberance debate. JPMorgan’s chief executive Jamie Dimon has repeatedly used the phrase “irrational exuberance,” borrowed from former Fed chair Alan Greenspan, to describe dealmaking activity that he says is running “gung-ho.”

Yet analysts at Fidelity note a structural difference from 2000: hyperscalers are largely funding AI capital expenditure from earnings rather than debt, keeping the capex-to-free-cash-flow ratio below 1, compared with nearly 4 at the dot-com peak, based on Fidelity’s bubble-indicator research. That distinction matters for systemic risk, since debt-fueled busts tend to transmit further into the banking system than equity-only corrections.

The Systemic Transmission Risk

Oliver Wyman’s analysis of a potential AI-led market collapse estimates that an equity crash on the scale of the early 2000s could erase approximately $33 trillion in value — more than annual US GDP — a scenario that would compound if financing tied to data-center and digital-infrastructure debt turns out to be more opaque than banks currently report, according to Oliver Wyman’s assessment of financial-sector exposure. US equity market capitalization currently sits at close to twice GDP, a higher multiple than at the dot-com peak.

Prediction markets have already begun pricing the risk. Polymarket data cited by Tekedia shows the probability traders assign to an AI investment-frenzy collapse by the end of 2026 climbing to 26%, up sharply in recent months as valuations in chip and hyperscaler stocks stretched further.

What Regulators Are Asking Institutions to Do

The BIS is not calling for a halt to AI development. Instead, it is urging financial institutions to build greater transparency into AI-related financing, particularly the private-credit channels that now fund a large share of data-center buildouts, and to stress-test balance sheets against valuation drops of 30%, 40%, or even 50% in AI-exposed equities. The Bank of England has separately warned that investors have not been adequately cautioned about downside scenarios tied to companies such as OpenAI, whose valuation more than tripled between October 2024 and the following year.

For markets in the UK, US, Singapore, and East Asia’s chip-manufacturing hubs, the message from regulators is consistent: the innovation is real, but the financing structure underneath it has not been fully stress-tested against a reversal in sentiment.

Discover more from The Economy

Subscribe to get the latest posts sent to your email.

AI

AI Bubble Risk 2026: BIS Warns Private Credit Could Trigger Financial Crisis

Published

1 week ago

July 3, 2026

Abdul Rahman

The Bank for International Settlements has told the world’s central banks something few wanted to hear in the middle of an AI-fueled bull run: the financing behind the boom now resembles the early architecture of a credit crisis. In its flagship Annual Economic Report, the Basel-based institution known as the central bank of central banks said that if AI returns disappoint and investors reassess risk, falling asset values combined with sudden funding withdrawals could transmit stress across the broader financial system, as first detailed by The Economy.

From Hyperscaler Capex to Systemic Fragility

The scale driving this concern is difficult to overstate. Microsoft, Amazon, Alphabet, Meta, and Oracle are collectively on pace to spend more than $1 trillion on AI infrastructure across 2025 and 2026 combined, a sum the BIS says already outpaces the group’s combined earnings and free cash flow. That gap is why hyperscalers have turned to debt markets at a pace unseen since the buildout of broadband infrastructure, with investment-grade bond issuance by major AI players exceeding $100 billion in six months, according to Oliver Wyman’s analysis of Dealogic and SIFMA data.

Fortune’s review of the BIS report frames the comparison in historical terms the institution itself invoked: the canal mania of the 1830s, Britain’s railway bubble of the 1840s, and the dot-com crash of 2000, each beginning with a genuine technological breakthrough that attracted more capital than commercial returns could ultimately justify, per Fortune. The BIS stops short of calling the AI boom a bubble outright, but its language leaves little room for comfort.

Private Credit’s Opacity Problem

The more acute concern sits outside public markets entirely. Private credit lending to AI companies surged from roughly $3 billion in 2010 to $40 billion last year, the BIS found. Because these loans flow through a web of investment funds, insurers, pension funds, and asset managers with little public disclosure, regulators cannot easily determine where losses would land if AI returns fall short. Unlike banks, these lenders have no deposit base and no central bank liquidity backstop, leaving forced asset sales as one of the few levers available if investors demand their money back.

That vulnerability is no longer theoretical. Blue Owl paused quarterly redemptions on a retail-facing direct lending fund earlier this year, an early sign of the liquidity strain described by Forbes. BlackRock’s TCP Capital Corp wrote down a private loan to an Amazon-seller aggregator to zero from full value, while bankruptcies at First Brands Group and Tricolor Holdings last September, each carrying billions in debt, have sharpened scrutiny of underwriting standards built during the ultra-low-rate years of 2020 and 2021.

Direct lending funds, an ecosystem now exceeding $1 trillion, have quadrupled their exposure to the AI and IT sectors over five years, and that exposure now represents about 15% of their portfolios, the BIS report notes. The Financial Stability Board, which monitors risk across 24 central banks, has separately warned that “significant data challenges” make the sector’s true exposure nearly impossible to map, with bank exposure estimates ranging anywhere from $220 billion to $500 billion depending on methodology, a spread detailed by IndMoney’s market analysis.

Why the Timing Is Especially Dangerous

The AI credit question is colliding with a second global shock that has nothing to do with technology. The closure of the Strait of Hormuz following the outbreak of the Iran conflict in February cut more than 10 million barrels of crude oil a day from global supply, a disruption larger than either the 1973 oil embargo or the 1979 Iranian revolution, according to the BIS report cited by Fortune. That energy shock has kept inflation risk elevated even as central banks weigh whether to ease policy, creating a scenario the BIS describes bluntly: the same monetary tightening needed to contain energy-driven inflation could be exactly what pops the AI-financed debt bubble.

Credit markets are already pricing in some of this tension. Spreads on bonds issued by AI-related companies rated BBB or higher have widened noticeably since the first quarter, briefly approaching a 20-basis-point increase in March, even as equity markets continue to price substantial further upside, a divergence flagged in the Economy’s coverage. Debt coming due from weaker private credit borrowers is projected to jump from $56.6 billion in 2026 to $215 billion by 2028, according to S&P Global data cited by IndMoney, concentrating refinancing risk at precisely the moment AI infrastructure utilization rates are becoming the market’s most important, and least verifiable, number.

What Happens if the Bet Doesn’t Pay Off

Not every analyst agrees the danger is systemic. The CFA Institute’s Enterprising Investor blog has pushed back on comparisons to the 2008 crisis, arguing that private credit’s structural mismatch is fundamentally different from the overnight funding of illiquid mortgage assets that caused the Global Financial Crisis, and noting that a well-diversified multi-strategy portfolio would likely be only marginally affected even by a serious AI correction, per CFA Institute.

But the BIS itself is not predicting collapse so much as demanding preparation. Its central recommendation is for what it calls “robustness” rather than the more fragile “resilience” the global financial system has shown so far, a distinction the institution says matters because a shock, whether a renewed inflation surge or a sharp AI-led repricing, could trigger a broader credit crunch. If half of the projected $6 trillion in AI capital spending through 2030 ends up debt-financed, the resulting credit buildup would exceed all broadband infrastructure investment since the birth of the commercial internet, Oliver Wyman’s modeling shows, and an equity crash on the scale of the early-2000s dot-com bust would, at today’s valuations, wipe out roughly $33 trillion in value, more than the entirety of US GDP.