UK finance says AI agents are widespread—but broad autonomy is still rare
A Bank of England and FCA survey found 80% of detailed-survey respondents reporting deployed AI agents, while only 2% reported broad autonomy. Third-party systems now dominate the most material uses, with thinner internal understanding.
The short version
- What happened
- The Bank of England and FCA received 196 responses: 101 larger firms and 95 smaller firms. Detailed findings normally come from 105 firms that completed the full survey, with percentages based on valid responses to each question.
- Who it affects
- For customers, the practical question is not whether a bank or insurer has an AI agent, but whether it can affect a payment, price, claim or credit decision—and who can explain and reverse that action when it is wrong.
- What remains unproven
- Descriptive, self-reported survey rather than an independent audit or outcome study
Keep in mind
Descriptive, self-reported survey rather than an independent audit or outcome study
196 total respondents, but most detailed results use the 105 full-questionnaire firms and valid responses vary by item
Changes in wording, scope and respondent populations limit direct comparison with 2024
No measured customer benefit, error rate, discrimination, incident loss or service-quality outcome
Responding regulated firms may not represent every UK financial-services business
Source analysis · 1 linked source. Primary analysed source date: 26 October 2026.
Read the full report · 4 min The evidence, methods, implications and limitations
Agents are common, but the consequential decisions remain constrained
The Bank of England and Financial Conduct Authority say AI adoption has moved quickly through UK financial services. Their 2026 survey received 196 responses—101 from larger firms and 95 from firms with fewer than 100 employees. A total of 105 firms completed the full questionnaire used for most detailed comparisons. Among those respondents, 86% reported using AI, up from 75% in the authorities' 2024 survey, although changes in questions and respondent populations mean the increase is not a clean panel comparison.
AI agents were reported by 80% of detailed-survey respondents, yet that headline covers very different levels of control. Eighty per cent reported decision-support agents, 46% constrained execution and 22% conditional autonomy; only 2% reported broad autonomy. Even after planned deployments were counted, no more than 9% expected agents to approve or decline customer outcomes, execute trades, initiate or block payments, or change operational limits. For a customer, that distinction matters more than whether a firm uses the word agent at all.[1]
Third-party dependence is growing faster than internal visibility
Third-party implementations accounted for 39% of all reported AI systems, up from 33% in 2024, and 53% of systems firms classified as high materiality, up from 35%. Concentration was stronger in infrastructure: the top three cloud or compute providers held 79% of reported relationships, while the top three model or API providers held 54%. These are relationship shares within the survey, not market shares for the whole UK financial sector.
The governance gap is concrete. Firms said they had deep knowledge of 77% of internally developed high-materiality systems but only 24% of third-party systems. For vendor-supplied systems, 81% instead relied on vendor assurance and contracts. Pre-deployment review, human approval and vendor due diligence were each reported by more than 90% of respondents, while ongoing monitoring, leakage testing and periodic review were each reported by only about half. Customer-impact metrics were monitored by 39%, and risk metrics by 34%.[1]
The distinction that matters
Three different claims hidden inside 'AI-agent adoption'
The authorities asked firms about deployment, autonomy and governance. Those measures should not be collapsed into one adoption headline.
Use
80% reported decision-support agents
Tools are widespread among detailed-survey respondents
Autonomy
2% reported broad autonomy
Independent high-consequence action remains rare
Oversight
24% reported deep knowledge of third-party high-materiality systems
Vendor dependence creates a visibility gap
The survey maps readiness, not whether AI helps customers
The report is descriptive and self-reported. Firms rated perceived benefits, risks and controls; regulators did not independently audit every system or measure accuracy, discrimination, complaints, losses or service quality. Percentages exclude firms that skipped each question, so the item-level denominator can be smaller than 105. Smaller firms could use a shorter questionnaire, and their responses are reported separately. Those design choices make the survey useful for a regulatory map, not a population estimate with sampling error for every UK firm.
What would change the assessment is outcome evidence: incident rates, customer redress, false approvals and denials, resilience failures and model changes tied to specific autonomy levels. Until then, the strongest conclusion is that agent-like tools have spread faster than broad autonomy, while material dependence on outside providers has grown. Regulators and firms now have a measurable gap to close between policies on paper and continuous evidence after deployment.[1]