UAE, UNFILTERED71% of DIFC Firms Use AI. Only 14% Have Deployed AI Agents
DFSA's 2026 AI survey shows 71% of DIFC respondents use AI, but only 14% have deployed agents. The gap between experimentation and governed operations is the story.
71% of responding DIFC firms use AI, but only 14% report agent deployment. Adoption is not equivalent to mature autonomous operations.
DIFC financial institutions, risk managers, compliance officers, AI vendors, regulators and banking customers.
Survey statistics cover respondents, not the entire UAE. Deployment maturity, vendor dependence and governance vary; bank trial productivity claims are not independently audited here.
Watch for detailed supervisory findings, governance progress, agent deployment rates and operational quality metrics in future DFSA reports.
Treat AI adoption statistics as survey measures. Ask what is genuinely deployed, who controls access and how decisions can be audited.
Artificial intelligence is now used by 71% of the financial firms responding to the Dubai Financial Services Authority’s 2026 survey. But just 14% of respondents say they have reached deployment stage with AI agents. Dubai’s financial centre has crossed an important adoption threshold without proving that autonomous systems are already running most of its banking operations.
The distinction comes from the regulator’s AI Survey 2026 findings, published on 2 November. The results cover responding DFSA-authorised firms in Dubai International Financial Centre, rather than every bank or fintech operating across the UAE. They chart a fast rise in adoption, while exposing a slower buildout of operational governance and full-scale deployment.
For Robius, the most useful story is not simply that AI adoption increased. It is that the same firms telling the regulator they are using AI are also reporting pilots, limited deployments, cloud dependencies and incomplete oversight arrangements.
Three years of growth, but not three years of equivalent deployments
| DFSA survey measure | 2024 | 2025 | 2026 |
|---|---|---|---|
| Respondents using at least one form of AI | 33% | 52% | 71% |
| Respondents using generative AI | 22% | 48% | 67% |
| Respondents expecting AI use to rise over three years | 54% | 75% | 84% |
These are proportions among survey respondents, not a verified count of all AI systems running in the market. The results should also not be turned into claims about the productivity of every firm using AI. An institution answering that it uses a generative model internally may have a very different risk profile from one that lets software approve transactions or alter customer records.
Among 2026 respondents using AI, 94% reported generative AI use. That reflects how mainstream the technology has become within the adopter group. It says nothing on its own about whether those tools are connected to live operational systems.
The most important gap is between using AI and deploying it widely
The regulator reports that 39% of AI adopters use applications across a considerable portion of a business area, compared with 23% in 2024. Yet 74% of adopters also say at least some AI applications remain at pilot stage or are used within only a small share of a business area.
These categories are not mutually exclusive. A large firm could use AI extensively for document review while keeping its new customer-service agent in a limited pilot. The presence of one mature application should not be mistaken for widespread maturity across the company.
That makes the much smaller agent number striking. More than half of respondents, 53%, reported some form of exploration, testing, piloting or deployment of AI agents, but only 14% reported reaching deployment. The regulator does not claim that every deployed agent can independently move money, approve financial products or operate without human intervention. Its survey reports a stage of adoption, not a universal technical capability.
Robius recently covered the DFSA’s separate research on agentic AI cyber resilience. That paper looks at speed, reach, authority and trust. The new adoption data offers the other half of the picture: how far firms have actually progressed, rather than only what their technology might eventually be capable of doing.
AI is mostly doing work inside institutions
Among respondents using generative AI in 2026, 91% reported applications in internal functions such as human resources, legal work or finance. Use in internal systems and controls, including audit, compliance and risk management, rose from 39% in 2025 to 52% in 2026. Client-facing or other externally consequential applications were less common, rising from 33% to 36%.
These are shares of relevant generative AI users reporting each category. They do not necessarily sum to 100%, because a firm can use AI in multiple places. The survey does not demonstrate that customer-facing financial decisions have become fully autonomous, and it would be wrong to read the 36% figure that way.
Why might internal uses grow faster? A summarisation or research assistant can be valuable without receiving authority to alter a customer’s balance or make an investment decision. Customer-facing systems require stronger safeguards around suitability, errors, disclosure, access and escalation. That is a practical inference from the different consequences of these functions, not a cause established by the survey.
Cloud dependence is rising faster than some governance measures
The DFSA reports that 75% of generative AI users relied on applications developed by third parties, up from 65% in 2025. Among all AI adopters, 73% said between 91% and 100% of their AI applications ran on cloud infrastructure, up from 62% the previous year.
That creates a useful question for boards and risk teams: can the firm identify which service would fail if a shared AI provider, model endpoint or identity platform became unavailable? A technology supplier’s service-level agreement is not a substitute for the financial institution’s own continuity plan.
Governance is developing but uneven. The DFSA found that 61% of AI adopters had an AI governance framework, 53% had organisational policies governing AI use, and another 28% said such policies were under development. Only 41% maintained an inventory of AI use cases, while 31% said they were developing one.
An inventory sounds mundane beside an agentic AI demo. It is also the basic answer to the question, “Which AI systems does the organisation actually depend on, who owns them and what can they access?” If an institution cannot answer that, it may struggle to supervise a tool that a business team deploys through an outside provider.
What does the 71% number not tell us?
The survey cannot tell a reader that 71% of all UAE financial institutions have adopted AI. It covers respondents to the DFSA survey among authorised firms in the DIFC. It also does not establish the economic return on AI spending, customer satisfaction, or whether any specific bank’s lending decisions are more accurate.
One commercial example illustrates the challenge. Emirates NBD told Emirates News Agency that about half of its technology budget had been allocated to AI projects over the past year. The bank also described a trial across 4,000 debt-collection accounts, split between AI and staff, and reported a 15% productivity improvement. These are the bank’s reported results. The interview did not provide a public independent audit of the trial methodology, customer outcomes or overall return on that budget.
These examples should not be blended into one performance score. The DFSA survey describes the wider respondent landscape; Emirates NBD’s case concerns a particular bank and a particular trial.
What financial customers should look for
Customers don’t need to know every model or vendor a bank uses. They do need clarity when an AI-powered service affects them: who is accountable for the outcome, how errors can be challenged, how sensitive information is handled and whether a real employee can review a disputed decision.
The Central Bank of the UAE issued responsible AI and consumer-protection guidance earlier in 2026, highlighting governance, fairness, explainability, human oversight and privacy. That guidance applies in the central bank’s supervisory context and should not be confused with the DIFC regulator’s particular survey statistics.
The next benchmark should be quality, not only adoption
Future DFSA surveys would be even more useful with consistent disclosure of the share of AI uses that are fully deployed, the incidence of failures and overrides, the proportion of high-impact decisions subject to human review and how institutions measure business value.
Those measures are harder than asking whether a firm uses AI. They are also closer to the outcomes that customers, boards and regulators actually care about.
The Robius Read
Seventy-one percent is a meaningful adoption milestone. But the real finding is the distance between experimentation and resilient, governed deployment. Only 14% of respondents have deployed AI agents, and the survey shows many institutions are still completing the foundational inventory and policy work.
Dubai’s next AI leadership test is not convincing financial firms to try the technology. It is proving that they can control it, audit it and measure the results once the pilots become essential infrastructure.
Sources and editorial method
- DFSA, AI Survey 2026 results, 2 November 2026
- Emirates News Agency, DFSA adoption and Emirates NBD statements, 2 November 2026
- DFSA, frontier and agentic AI cyber resilience research, 30 October 2026
- Emirates News Agency, Central Bank consumer AI guidance, 23 February 2026
Editorial note: Survey results are attributed to the regulator and distinguish all respondents from subsets of adopters or generative AI users. Bank trial results are separately identified as a company claim. No independent testing of institutions, models or outcomes was undertaken. Checked 5 November 2026.
