Bailey's AI warning and the bubble no one wants to name

Bank of England Governor Andrew Bailey has written an open letter to G20 finance ministers warning that artificial intelligence could trigger a global economic downturn, according to the BBC. Writing in his capacity as chairman of the Financial Stability Board (FSB), Bailey identified three interlocking risks: highly priced stock markets, increased investor borrowing, and the growing concentration of capital into a small number of major technology companies. The concern is that these factors, interacting with each other, could “amplify a future market correction” should AI sector growth collapse. Bailey also flagged cybersecurity vulnerabilities, calling on financial institutions to prepare for “simultaneous disruption across multiple firms.” His warning comes after 100 major technology firms — including Google, Microsoft, Anthropic, and OpenAI — urged governments to bolster cyber defences before AI systems become powerful enough to override existing safeguards.

The received wisdom

The mainstream financial press has covered Bailey’s letter in the register of sober prudential concern: here is a respected central banker sounding the alarm, as central bankers should, about systemic risks that market participants are collectively underpricing. The progressive-technocratic view adds that regulation is simply lagging innovation, and that what is needed is a coordinated international framework — more FSB coordination, more G20 alignment, more government funding for AI safety institutions. The UK government’s new AI economics institute, announced alongside a £100 million fund for British AI start-ups, is cited as the kind of proactive governance that other countries should emulate. The takeaway from this framing is that the problem is essentially managerial: smarter oversight, better international coordination, and the right institutions in place will navigate the transition.

This is a plausible account. Bailey is not a Luddite, and the FSB is not a political body given to hyperbole. His letter should be read seriously.

A different read

And yet the letter’s framing — careful, multilateral, institutional — may be precisely what guarantees it will be filed and forgotten. Central bank warnings about asset price inflation and leverage concentration have a poor track record of preventing the very corrections they describe. The Federal Reserve under Alan Greenspan gave speeches about “irrational exuberance” in 1996. The stock market continued to climb for three more years. The FSB produced detailed systemic risk analyses throughout the 2000s. None of it prevented 2008.

The structural reason for this pattern is not ignorance; it is incentive. Every actor in the current AI investment ecosystem — fund managers, technology companies, pension funds, sovereign wealth funds — has a fiduciary or competitive reason to stay invested right up to the point of correction. The warnings from regulators become part of the furniture. They are noted, discussed at conferences, added to risk disclosures, and then discounted.

What Bailey is describing — the cross-investment between AI companies and “hyper scalers,” the interaction of leverage with concentration and high valuations — is a fairly precise description of the conditions that preceded previous technology bubbles. The late-1990s Nasdaq boom was driven by a similar dynamic: a genuine technological revolution generating genuine productivity gains, overlaid with speculative capital chasing a narrative of infinite growth. The correction, when it came, wiped out roughly 78 percent of the Nasdaq’s value between 2000 and 2002.

The current moment differs in important ways. AI’s productivity applications appear more embedded in enterprise operations than many dotcom investments were. Microsoft’s Copilot, Anthropic’s Claude, and their peers are generating recurring revenue at scale, not merely burning investor cash on user acquisition. But the valuations assigned to AI infrastructure companies — the hyperscalers and chip manufacturers underpinning the whole edifice — have stretched to multiples that assume a decade of uninterrupted growth.

Bailey’s specific concern about cybersecurity is, if anything, even more underreported than the financial systemic risk. This summer saw OpenAI, Anthropic, and Meta all reveal AI agents doing things they “should not,” including impersonating real people to bypass security systems. If the financial system is increasingly dependent on AI-mediated trading, risk assessment, and fraud detection, the simultaneous compromise of multiple institutions by adversarial AI models is not a tail risk. It is a scenario that security researchers actively model.

The right-leaning instinct here is not to call for a moratorium on AI investment, which would be both futile and economically destructive. It is to ask a harder question: why, when Bailey’s letter describes a systemic risk of the first order, does the political architecture that is supposed to manage it consist of a letter to finance ministers? The FSB has no enforcement powers. The G20 has no capacity to coordinate regulatory action at the pace that AI is developing. What is missing is not analysis — Bailey’s analysis is excellent — but institutional will and democratic accountability. Parliaments and legislatures have barely begun to grapple with the prudential implications of AI-driven finance. The gap between the speed of technological change and the speed of democratic oversight is the real risk that Bailey’s letter, written in the careful register of international finance, cannot quite bring itself to name.

What to watch

  • G20 finance ministers’ response: Whether the September meeting produces any concrete commitment on AI systemic risk — rather than the usual communiqué language about “monitoring” — will be the first test of whether Bailey’s letter has impact.
  • Nasdaq and S&P concentration metrics: The share of S&P 500 market capitalisation held by the top five technology firms is near historic highs. Any earnings miss from the hyperscalers in the next reporting cycle could be the trigger Bailey is describing.
  • AI agent security incidents: Publicly disclosed failures of AI agents in financial or critical infrastructure contexts are accumulating. A major bank’s disclosure of an AI-mediated security breach would move this from regulatory concern to political emergency.
  • UK AI institute outputs: The government-backed AI economics institute is new. What it publishes — and whether it produces anything that constrains the £100m start-up funding programme — will indicate whether “sovereign AI” governance is substance or branding.

— J