DFSA Survey Finds 71% AI Use Among Respondents, but Agent Deployment Is Only 14%
DIFC finance firms report much greater AI adoption, but agent deployment, use-case inventory and oversight remain less mature than the headline numbers imply.

Illustration of the gap between AI adoption and production deployment. Not a screenshot of DFSA survey data. · Arabian Current / original digital illustration
The DFSA's 2026 AI survey reports AI use by 71% of respondents, but deployment of AI agents by only 14%. Usage is not equivalent to mature autonomous operations.
Artificial intelligence use is rising rapidly among financial-services firms responding to the Dubai Financial Services Authority's annual survey. But the regulator's 2026 results show that installing or testing AI is not the same as placing autonomous systems into routine financial operations. The distinction becomes clearest when the survey separates general AI adoption from agentic AI deployment.
According to the DFSA's 2 November release, 71% of survey respondents reported using at least one kind of AI in 2026, compared with 52% in 2025 and 33% in 2024. These are self-reported figures from respondents, not an audited census of all DIFC firms. Generative AI use was reported by 67% of respondents, up from 22% in 2024.
The agent adoption gap
The survey says 53% of respondents were exploring, testing, piloting or deploying AI agents. Only 14% reported having reached deployment. Combining the experimental and operational categories into one number would greatly overstate the maturity of the technology in day-to-day regulated finance.
The difference is visible in conventional AI too. Nearly three-quarters of AI adopters said at least some of their applications were still being piloted or used in only a small part of a business area. At the same time, the share of AI adopters reporting deployment over a considerable part of a business area rose from 23% in 2024 to 39% in 2026. Firms are expanding systems while retaining many limited deployments.
Where generative AI actually runs
Use of generative AI remains concentrated inside organisations. Among the users of generative AI surveyed, 91% reported internal applications such as HR, legal or finance. Use in internal systems and controls, including audit, risk and compliance, climbed to 52% in 2026. Use in activities affecting clients or outside stakeholders was lower, at 36%. The percentages describe overlapping use categories and should not be added.
This matters because the consequences of a drafting assistant differ from those of an agent that can alter account records or send instructions to another system. External access, delegated authority, approval boundaries and audit trails become more consequential when an AI tool can do more than suggest text.
Governance is still catching up
Among AI adopters, 61% reported having a governance framework. Only 41% reported maintaining an inventory or register of AI use cases, and a further 31% said one was being developed. An inventory is not proof of safety, but it makes it easier to know which tools are in use, the sensitive data they touch, their owners and how an institution would contain a failure.
Firms also rely increasingly on third parties. The DFSA found that 75% of generative-AI users depended on applications developed by outside providers, compared with 65% in 2025. Among AI adopters, 73% said almost all their AI applications ran on cloud platforms, up from 62% a year earlier. Outsourcing technology does not outsource the need to understand concentration risk, service recovery and data handling.
Why agent security looks different
A separate October paper from the DFSA and the Cambridge Digital Innovation and Regulation Initiative examines how advanced and agentic AI affect cyber resilience. It frames the practical issues around speed, reach, authority and trust. The paper is research, not a new binding set of supervisory rules, but it illustrates why firms need to understand the systems and permissions around their models.
Speed asks whether a harmful action can be contained promptly. Reach concerns the indirect suppliers and shared services on which an operation depends. Authority asks what the AI is allowed to do; trust concerns traceability and the ability to restore critical functions. Those questions are different from asking whether a model writes a plausible answer.
What the next evidence needs to show
The next useful benchmark is not simply a higher share of respondents saying they have tried AI. It is documented deployment at scale, effective monitoring, clear human intervention paths, measurable operating outcomes and tested third-party recovery procedures.
The DFSA says it will use the survey results to inform risk-based supervision of AI, technology and cyber exposure. Its numbers describe the state of the respondents at the time of the survey. They do not confirm productivity improvements or certify any individual firm's control environment.
In short, AI adoption among DIFC financial-services respondents has accelerated, but the evidence for operationally mature AI agents remains comparatively limited. That gap is now the more important question for boards, regulators and customers.
The regulator's survey found 33% reporting AI use in 2024.
The 2026 share reached 71%, with rising dependence on third-party applications and cloud.
Deployment evidence, use-case inventories and reliable oversight will be more telling than adoption alone.
What supports the signal.
Claims stay source-linked, with limits recorded.
Survey percentages refer to different respondent subgroups. The separate cyber paper is context and does not impose new requirements.
Operational use matters more than launch-day claims or model marketing.
Models, infrastructure, research and governance are separated when they imply different things.
The story is framed around what changes for UAE organisations and decision-makers.
Open the evidence record.
2 linked sourcesSupport + limitations recorded.
01Core evidenceNew DFSA AI Survey 2026 findings: AI adoption among DIFC financial services firms has more than doubled since 2024
Dubai Financial Services Authority
+
New DFSA AI Survey 2026 findings: AI adoption among DIFC financial services firms has more than doubled since 2024
Dubai Financial Services Authority
Provides AI adoption, agent deployment, governance, internal use and external dependency figures for 2026 survey respondents.
Responses are self-reported and category denominators differ; they are not a census or external audit.
02ContextDFSA and C:>DIR publish new paper on cyber resilience implications of frontier and agentic AI
Dubai Financial Services Authority
+
DFSA and C:>DIR publish new paper on cyber resilience implications of frontier and agentic AI
Dubai Financial Services Authority
Explains the speed, reach, authority and trust dimensions of agentic AI operational resilience.
The paper explicitly does not create new supervisory obligations or independently verify survey claims.
Evidence is the next test.
The DFSA's next annual survey and concrete evidence of sustained agent deployments, incident processes and control effectiveness.
