Bank of England and FCA survey: 80% of UK financial firms use AI agents, but only 2% give them broad autonomy

Global reference

• Country: United Kingdom
• Authorities: Bank of England and Financial Conduct Authority (FCA)
• Publication: Artificial intelligence in UK financial services – 2026 (fourth survey)
• Key date: Published 26 October 2026
• Sample: 196 respondents (101 larger firms, 95 smaller firms); 105 completed the full survey

AI agents: wide use, limited autonomy

• Of respondents with agents deployed, 80% use decision support, 46% constrained execution within set limits, 22% conditional autonomy and 2% broad autonomy.
• Even including plans for the next 12 months, no more than 9% expect agents to initiate, execute or block payments, or to change limits, thresholds and controls.
• Liquidity and treasury decisioning is an early area: 10% use AI there today, 17% including plans.

Growing reliance on vendors

• Third-party implementations made up 39% of reported AI systems, up from 33% in 2024 and 17% in 2022. For high-materiality systems the share was 53%.
• The three largest cloud providers held 79% of reported cloud relationships, and 98% of respondents use at least one of them.
• 37% of cloud relationships were rated both highly critical and hard to substitute. Only 35% of those had tested exit plans.
• 54% find it a challenge to apply model risk management to vendor-supplied models.

Controls and risks

• 94% have firm-wide AI policies. Human approval, pre-deployment review and vendor due diligence are each above 90%, but ongoing monitoring is only around half.
• 88% expect frontier AI cyber capabilities to have a moderate or high impact on their risk profile over the next two years.
• None of the listed benefits was widely seen as more than moderate today.

What to do now

• Classify every AI agent you use by the survey’s four autonomy levels, and record who approves its actions.
• Keep payments and changes to limits or controls out of agents’ reach unless a person approves each action.
• Test an exit plan for each AI or cloud provider that is critical and hard to replace.
• Add ongoing monitoring of AI outputs, not only a review before launch.

Our take

The headline is not 86%. It is 2%. Firms use agents widely but keep people in charge, which is the right order. The weak spot is vendors: most firms rely on contracts rather than real understanding of the models they buy, and few have tested how to leave.

Important. This summary is for general information and is not legal, regulatory or financial advice. It is based on public Bank of England and Financial Conduct Authority publications and may be incomplete or out of date. The official publications prevail; check the official source before acting.