AI use is spreading across UK financial services, but the Bank of England and FCA’s latest survey finds most deployed agents still support human decisions rather than act on their own. The useful distinction: adoption is moving quickly, while the most consequential forms of autonomy remain uncommon.
Watch Desk analysis
What happened
The authorities’ 2026 survey says 86% of respondent firms use AI, up from 75% in 2024. Generative AI now accounts for 56% of reported AI systems, compared with 17% for foundation models in the earlier survey.
Four in five respondents report AI agents in production, and another 14% plan to deploy them within a year. But 80% of firms using agents describe them as decision-support tools, where people choose and carry out the action; just 2% report agents with broad autonomy.
The report also finds third-party systems make up 39% of reported AI systems, up from 33% in 2024. They account for 53% of high-materiality systems, compared with 35% previously. These are survey findings, not a census: 196 firms responded, and most figures are based on the 105 that completed the full questionnaire. The Bank and FCA caution that comparisons with 2024 need care where questions, definitions or respondents changed.
Why it matters
The headline adoption rate can make AI in finance sound more hands-off than it is. The survey suggests firms are using agents widely, but mostly to advise people rather than approve loans, execute trades or move money. Those higher-stakes uses remain limited even when planned deployments are included.
Meanwhile, reliance on outside AI providers is growing, particularly among systems firms classify as high materiality. That makes vendor oversight and operational resilience part of the AI story, not a footnote for the procurement team.
Our read
This is a more useful picture than either “the robots run the banks” or “nothing has changed”. AI is becoming routine, while meaningful autonomy remains the exception. The awkward bit is that firms report stronger controls before deployment than after it: ongoing monitoring and periodic review are reported by only around half. The dashboard is filling up; keeping watch on what is already running is the less glamorous job that matters.
What to watch
- Whether agent deployments move beyond decision support into actions with direct customer or financial consequences.
- How firms monitor third-party systems after deployment, not just before purchase.
- Whether future surveys show changes in high-materiality use as well as headline adoption.
Discussion spark: Should financial firms be allowed to deploy broadly autonomous AI agents only after proving they can monitor them continuously, or is human approval the safer default?
Sources and evidence
- Artificial intelligence in UK financial services – 2026 – Bank of England (26 October 2026, 10:24 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.