AI agents in UK finance: wide use, little autonomy
The Bank of England and FCA found 80% of responding UK financial firms run AI agents, but only 2% give them broad autonomy. The agents recommend and people decide.
AI agents are common in UK finance, and almost none have broad autonomy. The Bank of England and the Financial Conduct Authority report that 80% of responding firms have AI agents deployed, while only 2% deploy agents with broad autonomy [1].
The numbers
- 80% of respondents have AI agents deployed, and a further 14% plan to within 12 months.
- 80% have agents in production that provide decision support, meaning the agent recommends and a person decides and acts.
- 2% have deployed agents with broad autonomy.
- No more than 9% expect agents to approve or decline customer outcomes, execute trades, initiate payments or change limits, even counting plans for the next 12 months.
- Use of AI overall is 86% of respondents, up from 75% in 2024.
On controls, more than 90% of firms report human approval, pre-deployment review and vendor due diligence. Ongoing monitoring, leakage testing and periodic review are each reported by around half.
What the study did
The Bank and the FCA ran their fourth survey of AI and machine learning in UK financial services and published it on 26 October 2026. A total of 196 firms responded, 101 larger and 95 smaller. Of those, 105 completed the full survey. The report gives percentages of valid answers to each question and does not state a response rate. It is a survey, not a census, and firms describe their own systems.
Our reading
In regulated finance, an agent is mostly a recommender with a human sign-off. The 80% and the 2% describe the same market. Adoption is wide because the human decision stays in place, and control is tight for the same reason.
That puts the risk in the sign-off. A person approving an agent’s recommendation is only a control if the person can see the evidence behind it and has reason to disagree. If the reviewer gets a clean summary and a deadline, the human step becomes a stamp.
The survey data shows where firms are exposed. More than nine in ten firms check an agent before it goes live. About half keep checking after. Agents change as the data, the vendor model and the staff around them change, so a launch review ages quickly.
For a board, ask for three things. Show the evidence file the human sees before signing. Show how often reviewers overrule the agent and what happened when they did. Show the monitoring results from the last quarter, not the sign-off from the launch.
For a regulator, the number to watch is the move from decision support to execution. Today at most 9% expect it. That shift will not come as a single announcement. It will show first as reviewers who stop overruling the agent.
Agents in finance are only as safe as the evidence the human signs.
Limits
The survey is self-reported, firms apply the report’s definition of an agent to their own systems, and 196 firms is a sample. The Bank and the FCA warn that 2024 and 2026 figures cover different firms and should be compared with caution.
We first posted this finding on X: https://x.com/EstimandLab/status/2115166346258641347
Sources
- Bank of England and Financial Conduct Authority, "Artificial intelligence in UK financial services 2026", published 26 October 2026, https://www.bankofengland.co.uk/report/2026/artificial-intelligence-in-uk-financial-services-2026 (checked 2026-10-29)