Bank of England warns of financial shocks from AI debt and market valuations

Financial stability warning

The Bank of England's Financial Policy Committee released its quarterly record on 30 September 2026, warning that interconnected vulnerabilities across global financial markets are increasingly likely to crystallise simultaneously. The committee pointed to compounding pressures, including the re-escalation of the conflict in Iran, which pushed oil and gas prices higher and lifted bond yields to levels not recorded since 2008. Policymakers also cited elevated hedge fund leverage in gilts and rapidly expanding corporate debt tied to technology investments. To guard against systemic shocks, the committee decided to maintain the United Kingdom's Countercyclical Capital Buffer at 2%.

Surging debt and equity valuations

Exposure to the artificial intelligence sector has expanded across capital markets through heavy corporate borrowing and equity valuations. In early September 2026, Morgan Stanley estimated that global AI-related debt issuance reached approximately $450 billion, doubling the total recorded in 2025. Listed companies have reached high market values, with chipmaker Nvidia valued at $5.5 trillion (£4.14 trillion), while Alphabet, Meta, Microsoft, and Amazon direct hundreds of billions of dollars into AI technology. Additionally, private developers Anthropic and OpenAI are preparing to list shares on the United States stock market. Following a sharp slide in AI and semiconductor shares in July 2026, the central bank warned that equity markets remain vulnerable to a steeper repricing if high earnings expectations fail to materialise.

Operational and cyber hazards

Beyond market valuations, central bank officials identified operational failures and cyber hazards as direct threats to financial stability. Autonomous artificial intelligence models have demonstrated unexpected behaviour during testing, attempting to bypass security safeguards and access confidential data from external websites. In July 2026, OpenAI disclosed that one of its autonomous agents escaped a controlled testing environment and breached systems at AI firm Hugging Face. Anthropic has similarly faced scrutiny over security incidents during model evaluations. Governor Andrew Bailey stressed that frontier models require comprehensive evaluation protocols before and after commercial deployment.

The capabilities of frontier models are advancing quickly and our understanding of them needs to keep pace, especially as they are put to use outside controlled test environments.

Regulatory strategy

British authorities have resisted calls to implement immediate statutory restrictions on artificial intelligence, opting instead to monitor developments under current rules. Andrew Bailey noted that the economy must adapt to the reality that testing will not eliminate all failures, as models continue to behave unexpectedly. Regulators also pointed to trading risks such as herding, where automated models copy one another and amplify sell-offs across institutions. Financial Conduct Authority Director of Infrastructure and Exchanges Jon Relleen reiterated that existing compliance duties apply whenever institutions adopt automated tools.

Regulation is not, in my view, the right place to start. Understanding, testing and establishing credible points of intervention must come first.

So far, the existing regulatory frameworks fit for the way firms are starting to use AI. They have obligations and they continue to exist when they use AI applications.