AI in UK financial services: looking beyond the numbers in the 2026 Bank of England and FCA survey
On 26 October 2026, the Bank of England and the UK Financial Conduct Authority (FCA) published the results of their fourth survey on AI in UK financial services. The survey updates the regulators’ evidence base on how firms are using, governing and managing AI, and supports their ongoing work against the backdrop of evolving AI capabilities and adoption.
Read alongside the FCA’s recent publications on AI (see our coverage here) and the final report of the Mills Review (see our coverage here), the results offer useful insights into the practical pressure points emerging as firms move from experimentation towards more widespread, and potentially more autonomous, deployment of AI.
In this blog post, we draw out the key themes from the survey and what they may mean for firms.
- AI governance is widespread, but ongoing oversight is less developed
Most respondents now report having formal AI governance arrangements: 94% have firmwide AI policies, principles or standards, and 84% have centralised AI or model-risk governance fora. But the survey identifies a notable gap between implementation of pre-deployment controls and ongoing oversight. Human approval, pre-deployment review and vendor due diligence were each reported by more than 90% of respondents, while ongoing monitoring, leakage testing and periodic review or recertification were each reported by only around half.
The contrast suggests that firms’ controls are more developed at the point of deployment than during subsequent use. As we noted in our recent briefing on the FCA’s good and poor practice, the FCA has identified oversight as an area where firms are still under investing; having a governance framework is only the starting point, and firms need to demonstrate effective oversight throughout an AI system’s lifecycle. This includes senior managers understanding the AI risks within their areas of responsibility.
- Third-party AI may be the harder governance problem
Third-party implementations now account for 53% of reported high-materiality AI systems (up from 35% in 2024). At the same time, firms reported less direct understanding of high-materiality third-party systems than internally developed systems, relying instead mainly on vendor assurance and contractual arrangements. Oversight of third-party AI was also the most commonly reported area of governance uncertainty or implementation challenge, particularly in relation to concerns about information security and/or operational resilience.
For firms, the practical challenge is to maintain effective oversight when their understanding of a system depends heavily on information supplied by a provider.
- Agentic AI is already here, but delegation remains limited
Although 80% of respondents reported having deployed AI agents (with a further 14% reporting plans to deploy them over the next 12 months), deployment remains largely assistive: 80% reported agents providing decision support, while only 2% reported agents with broad autonomy. Firms remain cautious about allowing agents to take consequential actions such as approving or declining customer outcomes, executing trades or placing orders, initiating, executing or blocking payments, or changing limits, thresholds and controls.
The distinction between assistance and delegation is important. As we discussed in our briefing on the Mills Review final report, the regulatory risk increases as AI systems move from supporting human action to taking action themselves. Moving from AI that supports human decisions to AI that takes action itself raises additional questions about accountability, consumer protection and redress. The Mills Review anticipates increasing pressure on existing frameworks for accountability, consumer protection and redress as autonomy increases. Firms will need to ensure that their governance and controls keep pace with the expansion in the use and autonomy of AI agents.
For now, the survey suggests that the widespread adoption of AI agents has not translated into widespread delegation of consequential decisions. It also provides a useful baseline against which the Bank of England and the FCA may monitor the transition.
- Cyber risk remains a leading concern and respondents expect it to grow
Respondents ranked data protection and security, third-party and supply chain risks, and risks specific to generative AI as the leading AI-related risks to firms today. On systemic risks, 57% rated their concerns about AI amplifying cyber risk as high or severe, rising to 80% over the next one to three years. Cybersecurity was also the highest-rated systemic risk in the 2024 survey.
These risks are likely to intensify as AI use becomes more widespread. The findings underline the importance of considering not only the risks arising from firms’ own AI systems, but also how AI may amplify the external threats they face. Firms’ systems and controls will need to respond to both.
- The emerging regulatory issue is less “what rules apply?” and more “how do we apply them?”
The survey illustrates the increasingly complex regulatory environment in which firms are developing and deploying AI. Respondents pointed not only to publications and guidance from the FCA and the Prudential Regulation Authority (PRA) (including the FCA’s AI Update and Consumer Duty and the PRA’s SS1/21 and SS2/21), but also to international frameworks such as the EU AI Act and NIST AI Risk Management Framework, underlining that firms are navigating an overlapping set of domestic and international requirements.
Importantly, the challenge is not simply the number of potentially relevant requirements. International regulation, including the EU AI Act where it indirectly affects AI adoption in the UK, was the second most frequently cited area in which firms reported uncertainty about best practice, and respondents identified similar uncertainty in applying the PRA, FCA and Bank of England’s expectations. This reinforces a broader theme in UK financial services: as AI adoption evolves, the challenge for firms is increasingly not simply identifying which existing rules apply, but determining what good compliance looks like when those rules are applied to new AI capabilities and use cases.
Recent publications setting out examples of good and poor practice, including from the FCA and Financial Stability Board, are an important step towards providing greater practical clarity on how existing regulatory expectations apply to AI. But this cannot be a static exercise: both regulatory guidance and firms’ approaches will need to continue evolving as AI technologies and their use cases develop.
- Looking ahead
The 2026 AI in UK financial services survey provides a useful snapshot of how AI adoption across the sector is evolving and the challenges that firms face. For firms, the priorities are to strengthen monitoring after deployment, maintain oversight of third-party systems and reassess controls as agents take on more consequential tasks. Practical regulatory guidance will help, but firms still need to demonstrate how their existing frameworks work for the AI they use.
