Global AI Markets
UK Financial Firms Use AI Agents Widely, but Only 2% Report Broad Autonomy
A 2026 Bank of England and FCA survey reveals the gap between agent deployment and independent action, and the controls businesses need.
What the Bank of England and FCA survey actually measures
The Bank of England and Financial Conduct Authority published their 2026 survey of AI in UK financial services on 26 October. Among responding firms, 80% reported having AI agents deployed, while another 14% planned deployment within 12 months. Yet only 2% reported agents operating with broad autonomy. These percentages refer to surveyed financial-services organizations, not all UK businesses.
The survey distinguishes decision-support systems from constrained execution, conditional autonomy and broad autonomy. Respondents commonly had agents that recommend actions while people remain responsible for decisions and execution. Agent deployment is therefore not interchangeable with a claim that the software independently runs the company.
The operational difference between agent capability and authority
The survey indicates that many companies can deploy an AI agent without allowing it to move funds, alter customer terms or execute trades without checks. That distinction matters in every sector: a system that drafts an invoice is not the same as one authorized to pay it; a system recommending a refund is not automatically allowed to issue it.
The Bank of England also highlights perceived risks around security, data protection, third-party dependencies and model-specific failures. Customers buying bespoke agents should ask for documented scopes, transaction-level limits, approval checkpoints and clear records of what the system suggested versus what it actually changed.
Regulation makes governance part of the product
The UK's Competition and Markets Authority updated its guidance on AI agents and consumer law on 30 October 2026. It explains that organizations remain responsible for consumer-law compliance when agents interact with customers, including where a third-party vendor supplies the system. The guidance covers transparency, monitoring, training and the need to address problematic behavior.
This does not mean every business AI task needs the same level of control. Internal summarization and externally binding actions carry different risks. For customer-facing workflows, an organization should specify identity disclosure, escalation rules, complaint handling, data retention, legal review and an emergency stop before granting more authority to the agent.
A deployment checklist for UK enterprises
Begin by mapping a concrete process: an incoming request, source systems, authorized read and write actions, exceptions, human approvals and a measurable end state. Then test the system against ordinary cases, adversarial inputs, unavailable APIs and incorrect model suggestions. Confirm that failed actions do not silently repeat or create financial side effects.
The survey demonstrates significant reported use within its sample, but it does not prove that broad autonomy is either necessary or beneficial. Companies should evaluate measurable cycle-time reductions, exception rates, customer outcomes, resilience and cost. The most commercially credible bespoke agent may be one that acts reliably within clear boundaries rather than one marketed as fully autonomous.
Sources and references
- Artificial intelligence in UK financial services — 2026 — Bank of England and FCA
- Complying with consumer law when using AI agents — UK Competition and Markets Authority
- Agentic AI in the UK: The Question Is How Much Authority We Give It — techUK
