AI Use in UK Finance Reaches 86 Percent as Agents Spread, Bank of England and FCA Find

Artificial intelligence has moved from an experiment to a routine part of work across much of the UK financial sector, according to a new joint report from the Bank of England and the Financial Conduct Authority. The survey, published on 26 October 2026, found that 86 percent of responding firms now use AI, up from 75 percent in 2024 and 58 percent in 2022. It is the fourth survey in the series, which began in 2019, and it draws on responses from 196 firms across banking, insurance, payments, lending and capital markets.

The headline figure tells only part of the story. Behind it sits a shift in the kind of AI firms run, a rapid rise in AI agents, and worries that have grown just as fast. Regulators stress one caveat: the survey reflects firms that replied, not every firm in the sector, and the respondent group differs from 2024. Even so, the direction of travel is hard to miss.

Generative AI now does most of the work

The biggest change since 2024 is the type of system in use. Generative AI now accounts for 56 percent of all reported AI systems, according to the report. That compares with foundation models accounting for 17 percent of reported use cases in the previous survey. In plain terms, firms have moved from testing language models on the side to building them into everyday tools for drafting, summarising, searching documents and answering staff questions.

Operations and IT remain the largest home for these systems, hosting about 31 percent of reported AI systems, up from about 22 percent in 2024. That makes sense when you think about where the volume sits in a bank or insurer: claims handling, customer queries, document checks, fraud screening and the long tail of back-office work that used to eat up staff hours. Customer-facing uses are growing too, but the report suggests the heaviest lifting still happens behind the scenes, where a mistake is easier to catch and a gain in speed is easier to measure.

Agents arrive, but on a short lead

The second striking finding concerns AI agents, systems that can carry out a sequence of steps toward a goal with limited human input at each step. The survey found that 80 percent of respondents already had AI agents deployed, with a further 14 percent planning to deploy them within the next 12 months. That is a fast move for a technology that was barely on most board agendas two years ago.

Yet the report draws a sharp line between having an agent and letting it act freely. Only 2 percent of respondents reported allowing agents broad autonomy. Most firms keep agents on decision support, where a human reviews the output before anything binding happens, or on constrained execution, where the agent can act only within tight limits. Around eight in ten firms use AI only for guidance or decision support in at least some settings, according to summaries of the findings.

Third-party systems are the growing risk

A third theme runs through the report, and it may matter more than the adoption numbers. Third-party AI systems now account for 39 percent of all reported AI systems, up from 33 percent in 2024. For systems firms rate as high materiality, meaning the ones whose failure would hurt most, the third-party share rises to 53 percent, up from 35 percent two years ago.

In other words, a growing share of the AI that matters most inside UK finance is built and run by someone else. That brings benefits, since few firms can build a frontier model alone, but it creates a dependency that is harder to inspect. The report notes that oversight of third-party systems remains a significant challenge, even though most firms say they have formal AI governance in place.

Related to that, firms pointed to cyber risk as a rising concern. Summaries of the survey put 57 percent of firms rating their current cyber risk as severe or elevated, with that share expected to climb to 80 percent within three years. Data protection and privacy rules in the UK, along with international regulation, were the most cited perceived regulatory constraints. None of this means firms plan to slow down. It means the next phase of AI in finance will be judged less on how many tools a firm has, and more on how well it governs them.

What this means for customers and staff

For ordinary customers, the survey points to a quiet change rather than a sudden one. AI is most likely to show up as faster answers, quicker document checks and smoother back-office handling, rather than as a chatbot making decisions about your money on its own. The low share of firms allowing broad agent autonomy suggests that, for now, the human in the loop is staying put where the stakes are high.

What to watch next

Three things are worth watching from here. First, whether the share of firms giving agents broader autonomy stays near 2 percent or starts to climb as controls improve. Second, whether third-party dependence keeps rising for high-materiality systems, and what firms do about exit plans and testing when a vendor changes its model. Third, whether the next survey shows the same gap between large and small firms, or whether cheaper tools and shared services help smaller firms catch up.

Frequently Asked Questions

What did the Bank of England and FCA survey actually find?

The joint survey, published on 26 October 2026, found that 86 percent of the 196 responding UK financial firms use AI, up from 75 percent in 2024. Generative AI accounted for 56 percent of reported AI systems, and 80 percent of respondents had AI agents deployed, while only 2 percent allowed agents broad autonomy.

Does the survey cover the whole UK financial sector?

No, and the regulators say so plainly. The findings reflect the firms that responded to the survey, not the entire sector, and the respondent group differs from 2024. The figures are best read as a strong signal about direction, rather than a census of every firm.

Should customers worry that AI agents are making financial decisions alone?

Based on this survey, that is not the norm. Most firms keep agents on decision support or constrained execution, with a human reviewing the outcome where the stakes are high. Customers are more likely to notice faster handling and quicker answers than to have a binding decision made by an agent without oversight.