TL;DR

The Bank of England and the FCA published their fourth survey of AI in UK financial services on 26 October. Across all respondents, 86% now use AI, against 75% in 2024, though the regulators caution that the two surveys’ respondents differ. Agents are common but tightly held: 80% have them in production for decision support, while only 2% let them act with broad autonomy.

The survey drew 196 responses, 101 from larger firms and 95 from smaller ones. Unless the report says otherwise, its figures come from the 105 respondents who filled in the full survey, and the regulators stress that this is evidence from participants rather than the whole sector.

Agents: everywhere, on a short lead

A further 14% of respondents plan to deploy agents in the next 12 months. 16% use agents for customer communications today and 34% including plans, against 8% and 25% for creating or changing records in core systems. Hardly anyone expects agents to take consequential actions on their own: even counting planned deployments, no more than 9% see them approving or declining customer outcomes, placing trades, moving payments or changing limits and controls.

Generative AI now makes up 56% of reported AI systems, and Operations and IT now host about 31% of them, against roughly 22% in the previous edition.

The third-party question

Another shift is in who builds the systems. Systems supplied by outside vendors now make up 39% of the total (33% two years ago) and 53% of the systems firms rate as high materiality, up from 35%. Concentration is heaviest in compute: the three biggest cloud and compute providers hold 79% of reported relationships, against 54% for the top three model and API providers.

Firms named information security and operational resilience as the main brake on third-party AI (40%), with limited transparency into vendor systems next (34%). Oversight of vendor AI was the governance area firms most often found difficult, and 54% picked applying model risk management to supplied models as a challenge.

Context

The survey series goes back to 2019, with editions in 2022 and 2024. It is worth reading alongside our coverage of the FCA chief’s Mansion House speech on market integrity. For a deeper look at the Bank’s stance, see our analysis of its right to intervene on AI standards.

Looking Forward

In our view, the third-party figures are the ones supervisors will act on. When more than half of high-materiality systems come from outside vendors, and three providers hold most cloud and compute relationships, the question shifts from how firms govern their own models to how they oversee someone else’s. For firms that buy AI rather than build it, vendor due diligence is likely to be where supervisory attention lands first.