THE NUMBERS, WITH SOURCES.

HOW TO READ THIS PAGE

Every figure below carries a trust rating, because the numbers on this subject are not equally solid and the difference decides how you may use them.

The distinction matters more than it looks. "97% are embedding AI agents" records what 70 UAE respondents told a survey run by a firm that sells AI advisory work. That does not make it false. It makes it a reading of sentiment, and sentiment moves faster than behavior, as the gap between the adoption table and the governance table shows.

ADOPTION AND DEPLOYMENT

GOVERNANCE AND CONTROL

Set against the adoption table, this is the most useful section on the page: near-universal reported adoption, and a minority who say they can enforce their own policy across it. Most of the control figures are global. Dataiku's CIO survey from September 2026 gives a UAE-only reading, and it points the same way: many agents in production, and very few CIOs who say they could contain a misbehaving one within two hours.

RETURNS AND BOTTLENECKS

GOVERNMENT AND POLICY

INFRASTRUCTURE AND INVESTMENT

WORLDWIDE COMPARATORS

WHAT IS NOT ON THIS PAGE

Some figures in circulation did not make it here, and the reasons are part of the method.

• Figures traceable only to a report its publisher has since corrected or withdrawn. A correction can remove the figure itself or the evidence behind it, so we leave such figures out until a primary source confirms them.
• A mandatory four-tier AI risk classification, a September 2026 self-assessment deadline and penalties of AED 10 million. Widely repeated, and not found in any statute. See our briefing on what actually applies.
• Figures reported under conflicting attribution. Where two sources are credited with the same number, we trace it to the publisher with a first-party record or leave it out. The 77% and 93% figures above are attributed to IBM's Race for ROI study and dated October 2025 for that reason; 2026 coverage often recirculates them without their original date. The 99% / 14% agentic AI gap quoted around the Dubai Chambers and Nasscom agreement failed the same test: The National credits it to a Ness Digital Engineering study of Indian companies, other coverage of the study gives 9–14% and names no country, and we could not find the study's method. It is no longer listed.
• Figures that circulated under the wrong publisher and the wrong geography. The 64% / 26% / 77% control figures above were briefly published here as UAE data from Cloudera. They are global figures from Check Point’s 2026 Cloud Security Report, reached through UAE-framed coverage that carried them without a geographic qualifier. Cloudera’s own UAE research reports a different 64% (scalability as an advantage of hybrid architecture) and a different 77% (readiness to manage agents), and no 26% at all. The rows now carry the correct publisher, date and scope. The same happened to the ~60% tool-access and 66% / 20% returns figures, listed here until October 2026 as Accenture Middle East data. They come from Deloitte's State of AI in the Enterprise, whose June 2026 Middle East release repeats the report's global figures as regional findings.
• Two events reported as one. A 10 June workshop in Dubai, attended by more than 300 participants from 50 federal entities, has circulated in August coverage as a revised attendance figure for the separate 9 August workshop that opened the program's strategic track. Both events are listed above, each with its own date.
• Market-size forecasts with no disclosed method. They are common in press coverage and of no use in a business case.

When we run a readiness audit, every figure in the report carries this same treatment: source, date, rating. A number that cannot be traced does not go in.