The Bank of England has issued a warning that artificial intelligence poses a potential risk to financial stability, adding regulatory weight to a debate that has been building across central banks, prudential supervisors and the financial institutions they oversee. The warning places the UK’s central bank alongside a growing cohort of regulators who view AI not merely as a productivity tool but as a source of new, poorly understood systemic exposure.
The core concern is structural. When AI models are used for forecasting, credit decisioning, liquidity management or trading strategy, and when those models share similar training data, architectures or vendor dependencies, correlated failures become more likely. A shock that triggers errors in one institution’s AI system may trigger similar errors across the sector simultaneously, amplifying rather than absorbing volatility.
Governance and explainability at the centre
Practitioners working at the intersection of AI and corporate finance have pointed to three specific governance gaps. First, many models in production remain opaque: outputs are acted on without the decision-maker understanding how the model arrived at a conclusion. Second, controls around model updates and retraining cycles are often informal. Third, the accountability chain between a finance team, a vendor’s model and a board-level risk committee is rarely clearly defined.
Explainability is not simply a regulatory preference; it is increasingly a prudential expectation. The PRA‘s supervisory statement on model risk management, published in 2023, requires firms to understand, document and challenge the models they use in material decisions. As AI models move from back-office analytics into front-office and treasury functions, that framework will apply with greater force.
The Bank of England’s warning also lands in a complex macro environment. Finance teams adopting AI for scenario planning and forecasting are simultaneously navigating geopolitical disruption, energy price volatility and an escalating cyber threat landscape. AI systems trained on historical data may be poorly equipped to model tail risks that have no close historical analogue, a limitation that becomes more consequential when the tool is trusted with capital allocation decisions.
Regulatory read-across
Beyond the UK, the regulatory trajectory is consistent. The EU’s AI Act classifies certain financial AI applications as high-risk, requiring conformity assessments, human oversight provisions and transparency obligations before deployment. The Basel Committee on Banking Supervision has flagged AI-related model risk as a supervisory priority. In the US, the OCC and the Federal Reserve have both issued guidance emphasising that banks remain accountable for decisions made by automated systems, regardless of vendor involvement.
For financial institutions, the practical implication is that AI adoption strategies will increasingly need to be designed with audit trails, human override mechanisms and documented model governance frameworks from the outset, rather than retrofitted after deployment. Vendors selling AI into the finance function will face growing pressure to support that documentation and to provide explainability tooling as a standard feature rather than an optional add-on.
The Bank of England has not yet published detailed guidance on AI-specific prudential expectations, but the direction of travel is clear. Institutions that build governance infrastructure now are better placed to meet whatever supervisory standards follow.
