Why AI Governance Will Define the UAE’s Global Leadership Ambitions

The UAE's bold AI vision demands transparency and trust, but cutting corners on governance risks costly fixes from preventable failures.

Paromita Gupta 10 hours ago

Topics

[Image: Chetan Jha/MITSMR Middle East]

Key Takeaways

Nearly 95% of UAE data leaders admit they lack full clarity on how their AI systems make decisions—even as adoption accelerates.

The UAE has the lowest perceived consequences of AI failure globally, which is fueling a governance paradox: high confidence in AI and low investment in explaining or auditing it.

Experts frame real AI governance as three questions, not one checkbox: can the system explain its decision, can that decision be reconstructed later, and can a human override it before it becomes final?

Global AI leadership won’t go to the country that moves fastest, but to the one whose AI systems can explain themselves first.

In 2023, a class action lawsuit was filed against UnitedHealth Group, an American insurance firm, alleging that its AI algorithm, nH Predict, routinely denied long-term care claims for elderly patients.

Two elderly patients on Medicare Advantage, a private alternative to Original Medicare—a 91-year-old man recovering from a broken leg and ankle, and a 74-year-old stroke survivor in rehabilitation—were allegedly discharged prematurely after an AI system terminated their insurance coverage.

The algorithm, trained on six million patient records, was intended to analyze a patient’s profile and estimate how many days of post-acute rehab people like them typically needed. It was expected to provide an estimate. What it ended up doing was converting statistical estimates into a hard operational target. The moment the algorithm said a patient “shouldn’t need care anymore,” staff were required to discharge the patient despite the treating physician’s recommendation to the contrary. The algorithm effectively replaced the physician.

The lawsuit did not argue that the technology was defective. It argued that the company had improperly relied on it to make coverage decisions that should have been assessed individually under the rules.

This raises a key question about AI usage: when algorithms make important decisions, who is really responsible—the machine or the humans who deploy it?

Research Context

Dataiku: Nearly 95% of UAE data leaders admitted they lack full clarity into how their AI systems make decisions, even as deployment accelerates. Only 17% of UAE leaders are actively asking their AI systems to “show their work,” while 62% lack confidence that their AI could pass a basic decision audit— despite 72% saying they’d still trust an AI agent to make autonomous decisions in critical business workflows.
IBM research: 80% of business leaders identified AI explainability, ethics, bias, and trust as major roadblocks to generative AI adoption.
Microsoft report: 59.4% of the UAE’s working-age population uses AI, placing the country at the top of global adoption rankings.

The UAE Is Deploying AI It Cannot Explain.

The UAE has made some of the most ambitious AI bets in the world. This goal is powered by over 10 national AI institutions and research centers, at least 9 sovereign‑AI and data‑infrastructure platforms led by groups such as G42, 7 dedicated startup and innovation clusters, and dozens of cross‑border partnerships with US tech giants such as Microsoft, IBM, Amazon, and Mastercard.

The country, along with Saudi Arabia, has emerged as one of the leading global markets for deploying agentic AI solutions. Safe to say, the UAE is working to become the world’s next tech hub.

However, that roadmap has its fair share of obstacles, including governance. While ambitions remain strong, a Dataiku report, based on 100 responses, highlights a stark reality at the ground level: nearly 95% of UAE data leaders admitted they lacked full clarity into how AI systems make decisions.

“Enterprises in the UAE, much like those globally, are betting on AI they don’t fully trust,” says Florian Douetteau, co-founder and CEO of Dataiku.

In February 2025, Abu Dhabi-based G42 published a formal AI safety framework that sets out protocols for risk assessment, governance, and external oversight and appointed a dedicated frontier AI governance board led by its chief responsible AI officer. The move made G42 one of the first AI companies in the region to formalize such an oversight structure.

This is no longer a technological problem, but a governance one, and the UAE leaders need to address it before a consequential failure comes knocking. The managerial implication is direct: leaders who scale AI deployment without parallel investment in the right mechanisms are not moving fast—they are accumulating governance debt.

When the perceived risk is low, it’s natural for organizations to deprioritize investments in governance frameworks.

— Gabriele Obino, vice president of Southern Europe & Middle East, Denodo

Perception vs Reality: The Governance Paradox

The recommendation to establish a governance structure may seem simple, but the challenge is largely structural and extends beyond the UAE. There is a global trust deficit in AI, even as companies and leaders pursue innovation. High-profile failures, such as Amazon’s recruitment AI bot, underscore this issue. These instances not only highlight the need for tech giants and leaders to improve their data to ensure AI remains unbiased and fair but also emphasize the importance of transparency in the decision-making processes behind these technologies.

Only 17% of UAE leaders actively ask their AI systems to “show their work,” while 62% show a lack of confidence in their AI systems’ ability to pass a basic audit of their decisions.

In scenarios like these, governance—the processes, standards, and guardrails that help ensure AI systems and tools are safe and ethical—needs to step up and take charge.

Despite the ambiguity, 72% of UAE data leaders would still trust an AI agent to make autonomous decisions in critical business workflows. Gabriele Obino, vice president of Southern Europe & Middle East at Denodo, says, “This surprisingly high trust level makes sense when you consider the early successes many organizations are seeing.”

First-mover advantage does work, with companies relying on “data flywheel”: early market entry yields more proprietary data and rapid algorithm refinement. “AI has delivered impressive efficiency gains and cost savings in controlled environments, creating real confidence in the technology. Add to that the competitive pressure to stay ahead in the market, and you can see why leaders are willing to move forward quickly,” Obino adds.

Once that foundation is in place, AI agents can begin to deliver meaningful impact. But they must operate within well-defined safeguards. The goal is not “AI everywhere,” but “AI within governance”.

— Mohammed Alkhotani, SVP & GM, Salesforce Middle East

However, this does not guarantee success in products and services.

Eager to be one of the first airlines to implement an AI-powered customer service bot, Air Canada launched the chatbot in 2022 without sufficient accuracy controls. The chatbot provided incorrect information to a passenger named Jake Moffatt, telling him that he could purchase a full-fare ticket for a family funeral and subsequently apply for a bereavement discount after the flight. When Air Canada denied his refund request, the airline argued that the chatbot was a “separate legal entity” responsible for its own actions.

Despite understanding the significance of AI governance, businesses are yet to pursue it actively.

While the UAE is one of the most proactive countries globally in establishing clear principles for responsible and trustworthy AI, it also has the lowest perceived consequences of AI failure. How might this influence the urgency or commitment to improving AI governance and traceability?

“When the perceived risk is low, it’s natural for organizations to deprioritize investments in governance frameworks. The focus shifts from deployment speed to building stable, trustworthy systems. While this might seem reasonable in the short term, it’s actually a risky approach,” says Obino.

This creates a governance paradox: because people think AI failures are rare, they feel less pressure to invest in explainability, documentation, and audits, even as they speed up AI deployment. For Douetteau, the UAE is betting on AI blindly, with a “sense of confidence that they won’t bear much consequence if things go wrong.”

The Show-Your-Work Framework

According to an IBM research, 80% of business leaders identified explainability, ethics, bias, and trust as major roadblocks to the adoption of generative AI.

“Governance must be embedded early, and establishing strong internal governance and clear accountability are key to success,” says Mohammed Alkhotani, SVP & GM at Salesforce Middle East.

While the instinct may be to treat governance as a single checkbox, leaders we spoke with suggest a narrower, more actionable approach: a set of three questions every organization should be able to answer before any AI system makes a decision.

Explainability: Can a system demonstrate, in non-technical language, why and how it reached the given output? Alkhotani points out that AI governance poses complex hurdles for enterprises in the region, as it lacks a single, comprehensive law or set of laws to provide guidance on deployment rules for organizations. “However, for any enterprise, the absence of a universal law is not an excuse for inaction and should be viewed as an even greater reason to instigate proactive, principles-based governance for AI,” he says.
Explainability lays down the foundation for everything that follows. But it is only as credible as the data underneath it. AI is only as strong as the data it relies on. “Implement a logical data fabric that provides a single, secure access point to all your data sources without requiring you to move or duplicate the data,” notes Obino. The provenance of a decision must be documented, and an active catalog that automatically tags data with business context, technical details, and classification must be maintained, rather than relying on static records.
Auditability: While explainability and auditability may sound similar, they are distinct concepts. If a third party requested that an organization reconstruct a specific decision, would it be possible? This reconstruction hinges on understanding who accessed what data and when. Implementing fine-grained, automated access control—tailored to user roles and data sensitivity—ensures that AI systems can access only the data they are authorized to use, thereby creating an inherent audit trail within the architecture. “The aim is to make governance an integral feature of your data architecture, rather than a manual checklist that your teams need to navigate,” Obino explains.
Human override: Does a person retain the authority and the practical means to intervene before the AI’s output becomes the final word? This is where United Health fell short — the algorithm’s estimate became a mandatory operational requirement that superseded the treating physician’s judgment.

“Once that foundation is in place, AI agents can begin to deliver meaningful impact. But they must operate within well-defined safeguards. The goal is not “AI everywhere,” but “AI within governance,” Alkhotani adds.

AI without oversight risks profound social and ethical harm, underscoring the vital role of governance. Frameworks strike a balance between innovation and safety, safeguarding human rights and dignity.

The more organizations focus on building responsible, transparent systems, the faster AI will move from hype to tangible business impact.

— Florian Douetteau, co-founder and CEO, Dataiku

“When you can’t explain how your AI reached a particular decision, customers, partners, and regulators naturally start questioning whether your automated processes are trustworthy,” says Obino.

“The encouraging news is that governance challenges can be overcome. The more organizations focus on building responsible, transparent systems, the faster AI will move from hype to tangible business impact,” adds Douetteau.

What Leaders Must Do Differently

Role	Action Required
C-Suite	Treat explainability as a board-level metric, not an IT checkbox. If a system cannot show its work, it should not be making decisions unsupervised. The UnitedHealth and Air Canada cases show how fast “the algorithm decided” stops being a defense and becomes a liability.
Managers	Managers using AI-assisted tools must know when a recommendation requires human intervention versus when it can run autonomously.
Data & Platform Teams	The foundation of the framework depends on a unified, secure data layer with active metadata management and policy-based access control, so that explainability and auditability have a reliable foundation rather than being bolted on after deployment.
Boards & Governance	Boards should ask whether AI deployment speed is matched by investment in auditability—not after an incident, but as a standing governance question. As the article notes, low perceived risk of failure is exactly what allows governance debt to accumulate unnoticed until it’s expensive.

Beyond the Adoption Race

As per a Microsoft report, with 59.4% of the country’s working-age population using the technology, the UAE currently leads the world in AI adoption. However, the UAE’s AI ambitions were never the problem — the gap between deployment speed and governance is. For Alkhotani, AI should be “deployed with confidence, delivering tangible business impact while maintaining the highest standards of responsibility, compliance, and trust.”

Low perceived risk made that gap easy to ignore; it did not make it disappear. The question every leader in this piece was really answering was not whether AI works, but whether anyone could explain it, audit it, or override it before the cost of not doing so came due.

Global AI leadership will not be won by the country that deploys AI fastest. It will be won by the one whose AI can explain itself first.

Topics

About the Author

Tags:

AI AI governance UAE

Topics

Share