The Trust Gap: Why AI’s Next Challenge is Convincing People to Believe It
As AI moves from experimentation to decision-making and increasingly autonomous systems, the UAE is making rapid progress on trustworthy AI. But for organisations to turn billions in AI investment into sustainable business value, trust, data quality and human oversight will need to keep pace.
Artificial intelligence has never had a bigger seat at the corporate table. Across the UAE, organisations are moving beyond pilots and proof-of-concepts and beginning to embed AI into the decisions, processes and customer interactions that keep their businesses running.
But there is an uncomfortable question sitting underneath that enthusiasm: what happens when people do not trust the decisions AI makes? The issue is becoming harder to ignore as AI moves from generating content to making recommendations, influencing decisions and, increasingly, taking action on behalf of employees.
For Michel Ghorayeb, Managing Director of SAS UAE, this is where the conversation around AI needs to mature. “We see that companies are moving beyond AI experimentation and looking more closely at what these investments are actually delivering,” he says. “Trust plays an important role in that, particularly as AI becomes part of business decisions.”
That connection between trust and business performance is becoming increasingly visible. The latest SAS research, with insights from IDC, found that organisations implementing trustworthy AI practices are 15 times more likely to report strong or high ROI from their AI projects. Globally, 62% of organisations with strong trustworthy AI practices reported strong or high ROI, compared with just 4% among those with weaker practices.
For Ghorayeb, the message is straightforward: trust should not be treated simply as an exercise in risk management. “Those with the strongest practices also achieve greater gains in revenue growth, cost savings and customer experience,” he says. “This tells us that trustworthy AI is not only about managing risk. It is about giving organisations greater confidence to use AI at scale, which is ultimately where its business value is realised.”
When AI makes a recommendation, who makes the final decision?
The challenge is that confidence does not automatically follow accuracy. An AI system may produce the right recommendation, but that does not necessarily mean an employee will accept it. The latest research found that 97.2% of respondents override AI-generated recommendations in at least some cases. The leading reason is not necessarily that the recommendation is wrong, but that people do not understand how the system arrived at it.
That creates an interesting paradox. Organisations are deploying AI precisely because they want machines to help them make decisions faster and, in many cases, better. Yet if employees routinely disregard those decisions, much of the potential value is lost.
“Human oversight will always be needed, and building confidence in AI starts with making its decisions understandable,” says Ghorayeb. Even an accurate recommendation can be difficult to accept when its reasoning is opaque. In the UAE, more than 40% of respondents identified insufficient explanation as the main reason for overriding an AI recommendation.
“We cannot simply expect people to trust and to follow AI outputs,” Ghorayeb says. The distinction matters particularly in industries where AI decisions can have a direct impact on people’s lives or finances. In financial services, for instance, AI is increasingly relevant to credit, risk, eligibility and financial crime detection.
In such environments, the objective cannot simply be to automate the decision. It is to create a system in which people can understand the recommendation, assess its relevance and intervene when necessary. “AI should support employees in making better decisions, not remove them from the process altogether,” says Ghorayeb. “This is especially true for critical industries and work processes.”
The UAE is moving quickly
If trust remains one of AI’s biggest challenges, the UAE is nevertheless making significant progress. The country’s Trustworthiness Index rose by 30.2 points in 2026 to 65.7, according to the SAS-IDC research, putting the UAE above the global benchmark. The improvement points to a broader shift towards more structured AI adoption and governance.
Ghorayeb attributes much of that momentum to the country’s increasingly deliberate approach to responsible AI. “The UAE has made a significant shift in how it approaches trustworthy AI over the past year,” he says. “The progress is being driven by a stronger focus on AI governance as adoption grows, supported by the country’s wider push around responsible AI.”
The UAE National Strategy for Artificial Intelligence 2031 has helped provide a framework for that development, while organisations across the public and private sectors are increasingly building governance capabilities around their AI deployments.
One of the most notable improvements in the latest findings was in Audit and Accountability, which rose 18.1 points year on year. For Ghorayeb, that improvement is important because it indicates that the country’s progress is not confined to policy discussions. “This suggests that progress is happening across several areas, rather than being concentrated only around governance,” he says.
AI is only as good as the data underneath it
There is, however, another familiar problem lurking beneath the AI conversation: data. Organisations may spend heavily on sophisticated models and increasingly autonomous systems, but the quality of their AI ultimately remains tied to the quality of the data feeding those systems.
The UAE findings reveal a sizeable gap between recognising that fact and actually doing something about it. Around two-thirds of respondents identify Data Quality and Governance as a top factor in AI reliability, yet only 13.5% say their organisations mandate data quality processes for every AI project.
More than a third still leave the responsibility to individual teams. That might work when AI is being developed in isolated pockets of an organisation. It becomes considerably more problematic when dozens or hundreds of AI projects are running across different departments.
“Recognising that data quality matters is much easier than applying the same standards across every project,” Ghorayeb says. “As AI expands across different teams and functions, organisations can end up with different approaches to how data is prepared, governed and checked.”
The result can be an organisation where one AI system operates against a completely different data standard from another. “Closing the gap means making data quality part of how AI projects are built from the beginning, with common standards and clear ownership of the data,” he adds. It is a less glamorous part of the AI conversation, but potentially one of the most consequential.
The arrival of the AI agent
The trust equation becomes even more complicated as organisations move from generative AI towards agentic systems. Generative AI can create content, summarise information and assist with decisions. Agentic AI potentially goes further, allowing systems to execute tasks and make decisions with considerably less human intervention.
That additional autonomy comes with an obvious consequence: people need to be more comfortable with the machine acting on their behalf. Globally, trust in generative AI stands at 76%, according to SAS research. For agentic AI, it falls to 66%.
The drop is not necessarily a rejection of agentic AI. Rather, it reflects the fact that the stakes change when AI moves from producing an answer to taking action. “As AI becomes more autonomous, organisations need stronger safeguards around what these systems are allowed to do, when human intervention is required and who remains accountable for their actions,” says Ghorayeb.
That does not mean every AI action needs a human approval button. Instead, organisations need to establish different levels of oversight based on risk. “The level of oversight should depend on the level of risk,” he explains. “A routine administrative task does not need the same controls as an action that could affect a customer or a financial outcome.”
This could become one of the defining governance challenges of the next phase of enterprise AI: determining where autonomy ends and human responsibility begins. “The aim is not to put a person in the middle of every action,” Ghorayeb says. “It is to make sure that as AI gains more autonomy, human oversight, explainability and accountability keep pace with it.”
Building the infrastructure for an AI economy
Trust and governance may dominate the headlines, but neither can function effectively without the infrastructure underneath them. The UAE has made substantial progress in building the data infrastructure needed to support its AI ambitions. Organisations are increasingly moving away from fragmented and siloed data environments towards more mature, managed infrastructure.
Yet the transition is far from complete. According to SAS, 17.5% of enterprises globally have fully optimised data infrastructure mature enough for the demands of agentic AI. Organisations with an optimised data foundation are four times more likely to expect strong ROI from AI projects.
The UAE figure cited by Ghorayeb is 17.6% of organisations reaching an optimised level of data infrastructure. “More autonomous systems place greater demands on the data behind them, so continuing to strengthen these foundations will be crucial,” he says.
It is an important distinction. The next generation of AI will not simply require better models. It will require better-connected, better-governed and more reliable data environments from which those models and agents can operate.
The human factor could be the biggest gap
There is one other foundation that organisations cannot afford to overlook: people. As AI becomes more capable, organisations will need employees who understand not just how to use these systems but how to question them, govern them and recognise when something has gone wrong.
Yet only about a quarter of UAE organisations are investing in AI ethics skills development, according to the findings cited by Ghorayeb. “The need to strengthen the expertise around how AI is deployed and governed” is becoming increasingly important, he argues.
That could prove just as important as investing in the technology itself. An organisation can have sophisticated AI models, extensive data infrastructure and detailed governance policies, but without people who understand how all three interact, those investments may fail to translate into meaningful business outcomes.
The next 24 months
For UAE organisations, the next phase of AI adoption may therefore be less about asking whether AI works and more about asking whether the organisation is ready to scale it. “The priority should be turning AI from individual projects into something organisations can scale and use consistently across the business,” says Ghorayeb.
That means getting the fundamentals right: strong data foundations, clear governance, explainability and the skills required to manage increasingly autonomous systems. The UAE has already established an ambitious direction for AI. The next challenge is translating that national ambition into sustainable value inside individual organisations.
“The country has set an ambitious direction for AI through its national strategy, and businesses now have an opportunity to translate that ambition into value,” Ghorayeb says. That may ultimately be the more important AI race.
The first phase was about getting organisations to experiment with artificial intelligence. The next is about getting them to trust it enough to use it at scale — without surrendering the accountability that comes with making consequential decisions. For businesses, that means the future of AI may depend as much on governance, data and people as it does on algorithms.
And increasingly, the organisations that understand that may be the ones that get the greatest return from the AI revolution.
