A fresh 2026 survey shows AI use soaring in UK finance, generative AI now the majority, and third‑party solutions reshaping risk management.
The latest questionnaire completed by 196 UK-based financial institutions paints a vivid picture of how artificial intelligence has moved from experiment to everyday tool. 86% of respondents confirm they are running AI workloads today – a jump from 75% just two years earlier – and the pace of change shows no sign of slowing.
Banks, both domestic and international, report universal adoption, while insurers sit just behind at 93%.
While the sheer number of AI programmes varies widely, the concentration is striking: almost half of the firms disclosed ten or fewer AI applications, yet the ten most-active participants own 43% of every reported system.
In production, 94% of respondents host at least one AI model, rising to 97% when pilots are counted, and to 99% when planned roll-outs for the next twelve months are added. Median production counts differ by sector, with international banks and broker-dealers operating a median of 43 models, UK banks 22, and insurers 18.
Generative AI eclipses traditional models
One of the most decisive shifts concerns the technology mix. In 2024, foundation models – including large language models – represented only 17% of all AI cases. By 2026, generative AI accounts for 56% of every system reported, making it the dominant paradigm across the industry. The surge is evident in every business line, but Operations & IT now host roughly 31% of all AI deployments – up from 22% two years ago – while Risk & Compliance and other functions each remain under 11%.
Use of generative tools is especially strong in the productivity arena. Over 86% of firms employ AI to automate routine coding, summarise documents or generate insights, and that share climbs to 91% when forthcoming projects are included. By contrast, sectors such as liquidity management, treasury decisioning and algorithmic trading still show modest adoption, indicating a still-nascent appetite for AI-driven market-making.
AI agents spread but stay under human control
Agentic applications have also taken off. Eighty percent of respondents already run at least one AI agent, while another 14% plan deployment within the coming year. Enterprise-wide and coding assistants dominate the current landscape, and multi-agent systems are expected to grow fastest. Despite the breadth of deployment, autonomy remains cautious: 80% of agents provide only decision-support – they suggest actions, but humans retain final authority. Constrained-execution agents appear in 46% of firms, conditional-autonomy agents in 22%, and fully autonomous agents in just 2%.
Customer-facing agents are still emerging; 16% of firms now use agents for automated communications, rising to 34% when planned projects are counted. Functions that would hand over critical authority – such as approving credit decisions, placing trades, or adjusting payment limits – remain rare, never exceeding 9% even with future intentions.
Third-party AI becomes a material component
Reliance on external providers has deepened markedly. Third-party implementations now represent 39% of all AI systems, up from 33% in 2024, and they form 53% of every high-materiality deployment – a jump from 35% previously. Cloud and compute services are highly concentrated: the three largest providers host 79% of all relationships, and 98% of firms use at least one of them. Model-API hosts show a similar pattern, with the top three covering 54% of connections.
Security and resilience dominate the list of constraints. Forty percent of respondents cite information-security or operational-resilience concerns as the primary barrier to third-party AI, while 34% point to limited transparency around training data, model design or failure modes. Contractual negotiation difficulties rank second when moderate constraints are considered.
Balancing benefits, risks and systemic exposure
When asked to rank the most tangible advantages, firms placed operational efficiency, automation and productivity at the top, followed closely by improvements in cybersecurity and resilience. Looking three years ahead, respondents anticipate the biggest upside in revenue generation, competitive positioning and further strengthening of cyber defences.
Conversely, the most pressing threats today centre on data protection and security, third-party supply-chain exposure, and the unique hazards of generative AI – hallucinations, prompt injection and inadvertent data leakage. Even though current risk scores are slightly lower than in 2024, participants expect every category to climb toward moderate-to-high severity within the next one-to-three years.
Systemic concerns mirror these trends. Over half of the sample rates cyber-risk amplification as a high or severe issue today, and 80% project this concern to intensify. Critical third-party dependencies also rise sharply, with the share rating them as high or severe moving from 40% now to 63% in the near future. Interconnectedness and the potential for AI-enabled market manipulation rank among the fastest-growing systemic worries.
Governance structures have largely kept pace. Ninety-four percent of firms maintain firm-wide AI policies, and 84% run a central AI-risk governance forum. Pre-deployment controls such as human approval, vendor due-diligence and model-review dominate, while post-deployment monitoring and periodic recertification sit at about 50% prevalence. Understanding of high-materiality models diverges by source: 77% of respondents rely on deep internal knowledge for home-grown systems, but only 24% feel the same about external models, opting instead for vendor assurances and contractual safeguards.
Finally, sustainability considerations are emerging but not yet mainstream. Fifty-seven percent of participants say they factor AI’s environmental impact into decisions, yet 82% do not disclose these effects publicly. Among those not yet reporting, more than a third intend to begin external reporting within two years.
