Bank of England Says It Is Testing AI Risks to Financial System: 7 Things to Know

1. The Bank says it is already testing risks, not taking a wait-and-see approach

Reuters reported that the Bank of England rejected the idea that it was taking a passive approach to AI risk. In Sarah Breeden’s letter to the Treasury Committee, the Bank said it was conducting scenario analysis and simulations to test how AI-related stress could affect the financial system.

That point matters because it answers the main policy criticism directly. The Bank’s position is that it is already studying how AI investment and adoption are changing the structure of finance, rather than waiting until a crisis forces a response.

2. The Bank does not think advanced AI is systemic in UK finance yet

The April 2026 Financial Policy Committee record is clear on the current baseline. The FPC said there was little evidence that financial firms had adopted more advanced forms of AI in a manner that would present systemic risk today.

It specifically noted there was little evidence that advanced AI was already making core financial decisions such as credit or insurance underwriting, or driving core trading and investment activity at scale. That is an important nuance. The Bank is warning about rising risk, not declaring that the UK financial system has already crossed the line.

3. The biggest forward-looking concern is agentic AI in payments and financial markets

The Bank’s most explicit forward-looking concern is agentic AI, meaning systems that can take more autonomous action toward a goal. In the April 2026 record, the FPC asked the Bank and the FCA to draw out risks from agentic AI deployment in payments and financial markets specifically.

That is a telling choice. Payments and markets are areas where speed, automation, and network effects matter a lot. If AI agents start handling more decision-making there, even small model failures or poorly aligned incentives could have outsized consequences.

The FPC’s wording is also important. It said firms’ private incentives to deploy agentic AI might fail to internalise negative externalities, such as more payments fraud or markets becoming more prone to sharp movements. In plain terms, what looks rational for one firm might still make the system less stable overall.

4. Herding behaviour is one of the clearest market risks

Reuters said the Bank’s testing will focus on herding behaviour that could amplify selloffs during periods of market stress. That lines up closely with the Bank’s April 2025 Financial Stability in Focus note, which warned that greater use of AI-driven trading and investment strategies could lead participants to take increasingly correlated positions.

The mechanism is straightforward. If many firms rely on the same models, the same data sources, or very similar system designs, they may react in similar ways when conditions change. In calm periods that can look like efficiency. In stressed periods it can turn into procyclical selling, fire-sale dynamics, and more violent market moves.

The Bank is especially concerned about system-level behaviour here. Individual firms may think they are managing risk well, but if they all respond in the same direction at the same time, the collective outcome can still damage financial stability.

5. AI service-provider concentration could become a systemic vulnerability

The Bank’s AI risk framework does not stop at models inside banks and funds. It also focuses on the infrastructure around them. In its 2025 Financial Stability in Focus report, the Bank warned that financial institutions often depend on a small number of outside providers for models, cloud compute, and data.

That creates a familiar resilience problem in a new form. If many firms rely on the same vendor-provided models or supporting infrastructure, outages or weaknesses at those providers could affect large parts of the system at once. The Bank specifically noted that some AI and data providers could emerge as potential future critical third parties as the sector’s reliance increases.

This is one reason the wider UK policy debate matters. Reuters reported that the Treasury Committee criticized the government for moving too slowly on bringing major AI and cloud companies into the Critical Third Parties regime. Even before those designations are made, the Bank is already treating concentration risk as a stability issue worth stress testing.

6. Cybersecurity cuts both ways

AI can help financial firms defend themselves, but the Bank is equally concerned that it can make attackers more capable. Reuters linked the issue to Anthropic’s Mythos launch and Andrew Bailey’s warning that the model may have opened up major cyber risks in new ways.

The Bank’s 2025 financial stability note says the effect is genuinely two-sided. AI can improve fraud detection, cyber defence, and operational monitoring. But it can also help malicious actors scale attacks, generate more convincing social engineering, exploit shared model weaknesses, or target vulnerabilities in third-party AI systems.

From a stability perspective, the problem is not only whether one bank gets hit. It is whether common AI tools, shared vendors, or coordinated attacks turn firm-level cyber incidents into broader operational contagion across the system.

7. More monitoring, more data gathering, and possibly more guardrails are coming

The Bank’s current stance is not a ban on AI in finance. It is closer to structured escalation. The FPC supported continued monitoring of AI adoption by regulated firms, a re-run of the joint Bank and FCA AI survey in 2026, more market intelligence gathering, and more work with industry on AI risk-management practices.

That sits on top of a broader evidence base the Bank has already built. In its 2025 innovation report, the Bank said 75% of survey respondents were already using AI and another 10% planned to use it within three years. It also said the AI Consortium would help explore risks such as dependence on third-party providers, the use of similar models across firms, and challenges around explainability and transparency.

The practical implication is that regulation may stay technology-agnostic, but it will not stay technology-blind. If adoption rises faster, or if the Bank sees evidence of growing correlations, operational concentration, or AI-enabled market instability, the current monitoring phase could turn into a more active guardrail phase.