---
id: "lib-2026-10-26-artificial-intelligence-in-uk-financial-services"
type: "library"
title: "Artificial intelligence in UK financial services - 2026"
lang: "EN"
org: "Bank of England; FCA"
source_type: "Survey"
date: "2026-10-26"
week: "2026-W44"
url: "https://nosov.ai/library/lib-2026-10-26-artificial-intelligence-in-uk-financial-services/"
original_url: "https://www.bankofengland.co.uk/report/2026/artificial-intelligence-in-uk-financial-services-2026"
tier: "A"
tier_name: "Strong evidence"
tier_reason: "The joint Bank of England and FCA survey of AI in UK financial services (196 firms), following the 2019, 2022 and 2024 rounds, with a stated method, definitions and comparisons to 2024; authoritative and repeatable, though based on firms' own reporting."
summary_basis: "full-text"
original_minutes: "35"
tags: ["business transformation", "ai agents", "financial services", "third party concentration"]
---

# Artificial intelligence in UK financial services - 2026

Tier A (Strong evidence): The joint Bank of England and FCA survey of AI in UK financial services (196 firms), following the 2019, 2022 and 2024 rounds, with a stated method, definitions and comparisons to 2024; authoritative and repeatable, though based on firms' own reporting.

Summary based on the full text.

## In short

The Bank of England and the Financial Conduct Authority published their 2026 survey of AI use in UK financial services on 26 October 2026, following earlier rounds in 2019, 2022 and 2024. Adoption keeps rising and has shifted to generative AI. Agents are common but almost all recommend while humans decide; very few firms plan to let agents approve customers, trade, move payments or change limits. Third-party systems now make up most high-materiality AI, cloud and model supply is concentrated, and controls are much stronger before deployment than after it.

## Key numbers

- 86%: respondent firms using AI, up from 75% in 2024
- 2%: firms with AI agents deployed at broad autonomy, against 80% using agents for decision support
- 39%: share of reported AI systems that are third-party implementations, and 53% of high-materiality ones
- 79% and 54%: share of reported relationships held by the top three cloud/compute and top three model API providers
- around half: firms reporting ongoing monitoring, leakage testing or periodic review, against over 90% for pre-deployment controls

## What it reports

- 86% of respondents use AI (75% in 2024, 58% in 2022). Generative AI accounts for 56% of reported AI systems. Operations and IT hold about 31% of systems; no other area exceeds 11%.
- 7% of systems are rated high materiality (16% in 2024), though the absolute number rose. Ten firms hold 43% of all reported systems and 62% of high-materiality ones.
- Agents: 80% have them deployed and 14% more plan to within 12 months. By autonomy: decision support 80%, constrained execution 46%, conditional autonomy 22%, broad autonomy 2%. Even with plans included, no more than 9% expect agents to approve or decline customer outcomes, execute trades, handle payments, or change limits and controls.
- Third-party implementations: 39% of systems (33% in 2024, 17% in 2022) and 53% of high-materiality systems (35% in 2024). For third-party high-materiality systems, 81% rely on vendor assurance and contracts; 24% claim deep knowledge of them, against 77% for in-house systems.
- Concentration: top three cloud providers hold 79% of relationships and 98% of firms use at least one of them; top three model API providers hold 54%. Of 169 cloud relationships, 37% are both highly critical and hard to substitute, and only 35% of those have tested exit plans.
- Constraints: information security and resilience is a major constraint on third-party AI for 40%, limited transparency for 34%. Integration into processes (23%) and legacy systems (22%) top the non-regulatory list; data protection (42%) and international rules (29%) the regulatory one.
- Governance: 94% have firmwide AI policies and 84% central governance fora. Human approval, pre-deployment review and vendor due diligence exceed 90%; ongoing monitoring, leakage testing and recertification sit at around half.
- 88% expect frontier AI cyber capabilities to have a moderate or high impact on their risk profile within two years; 64% say vulnerability management and patching got the largest increase in management attention.

## How they got it

Survey of 196 regulated firms (101 larger, 95 smaller); firms with fewer than 100 staff could choose a short form, and 105 completed the full questionnaire. Most figures use full-survey responses, with percentages over valid answers to each question. Results are anonymised and grouped into six sectors. Definitions of AI systems and agents follow the OECD.

## Limits

- Self-reported by firms that chose to respond; the authors say it is not a full picture of the sector.
- The respondent mix and question wording changed since 2024, so trend comparisons need care (for example, "GenAI systems" in 2026 against "foundation model use cases" in 2024).
- Some findings rest on small subsets: 48 firms on vendor update controls, 36 on data providers.

## Why it matters for companies

This is one of the few regulator-run, repeatable counts of agent autonomy in a heavily controlled industry, and it shows where the line sits in practice: agents draft, summarise and recommend; humans still approve money, credit and limits. Two gaps apply well beyond banking. Most material AI now comes from vendors that firms understand mainly through contracts, and post-deployment monitoring lags pre-deployment review by a wide margin. Any company running vendor models should check whether it can test vendor updates, roll them back, and exit a provider it now depends on.

Original: https://www.bankofengland.co.uk/report/2026/artificial-intelligence-in-uk-financial-services-2026