UAE Leads in AI Agents, but Few Can Stop a Rogue One Fast: 6 Things to Build
Zeta42 · 29 September 2026
UAE firms run more AI agents than any other market Dataiku surveyed, yet only 5% of CIOs can contain a rogue one within one to two hours. Six things to build in early.
What the Dataiku survey found
UAE companies run more AI agents in production than their counterparts in any other market surveyed by software firm Dataiku, Gulf News reported. The study, conducted by Harris Poll across eight markets, found that 62 per cent of UAE chief information officers believe their organisation has more than 50 agents running in production. Fifteen per cent said they operate more than 500, the highest share of any country surveyed and well above the 9 per cent global average.
UAE teams are good at spotting trouble. Eighty per cent of UAE CIOs said they had encountered an agent that violated business intent, policy or expectations while still operating within its technical parameters. But 62 per cent said the incident was caught before it caused customer, financial, compliance or operational impact, the highest share globally against 48 per cent overall.
Stopping an agent is the weak point. Only 5 per cent of UAE CIOs said they could reliably identify and contain a problematic agent across environments within one to two hours, half the worldwide figure of 10 per cent. Another 27 per cent could act later the same day, so more than two-thirds could not reliably contain an agent on the day it was found. Only 18 per cent have a fully unified view of agents across platforms, data environments and business applications.
"Detection and control are different capabilities, and the research shows a real gap between them," said Florian Douetteau, CEO and co-founder of Dataiku. An agent you can spot in one system but cannot stop everywhere else, he said, "isn't contained". It has "just been noticed".
Six things to build in from day one
For any UAE organisation putting agents into production, the lesson is that the controls for stopping an agent are far easier to build in at the start than to add later across dozens of them. These are the six we would insist on.
- Keep one register of every agent. You cannot contain what you cannot find. Record each agent's owner, purpose, the systems it touches and the model it runs on. Sid Bhatia, Dataiku's Area Vice President and General Manager for the Middle East, Turkey and Africa, called for a unified view of agent populations, and only 12 per cent of UAE CIOs said agent lifecycle management is standardised across their organisation.
- Write limits in business terms as well as technical ones. The 80 per cent figure describes agents that broke business intent, policy or expectations while still operating within their technical parameters. System permissions alone will not catch that. Set the rules the business cares about, such as spending caps, which counterparties an agent may deal with and which actions always need sign-off. The agents Zeta42 builds work within limits the client sets, and every action they take is logged.
- Log every action. A full record of what each agent read, decided and did lets someone spot a problem early and trace its cause afterwards. Without it, even a caught incident is hard to explain.
- Build a stop switch that works everywhere, and name who may use it. This is the gap the survey exposes. Each agent needs a way to be paused, or have its access revoked, in one step across every system it touches. Bhatia called for clarity on who has authority to act and the ability to intervene immediately, regardless of where an agent was built or which system it uses. Test the switch before launch, not during the first incident.
- Send the unclear cases to a person. An agent should know when to stop and ask. In one Zeta42 example, an agent checks an invoice against the purchase order and the delivery note, and asks a person only when the three do not agree. Routine work flows through, and a human reviews the exceptions.
- Plan for the model going away. According to Gulf News, 63 per cent of UAE CIOs plan to use more or different models to strengthen business and AI continuity, and 62 per cent are considering open-source alternatives as a hedge against future unavailability. Design agents so the model underneath can be swapped without rebuilding the rest.
Why this gets skipped
The pressure on technology leaders runs one way. All UAE CIOs surveyed reported greater board pressure than in 2025 to show measurable return on AI investment, and 93 per cent said their CEO has indicated their job security depends on AI outcomes. Under that kind of pressure, registers, logs and stop switches look like overhead. They are what lets a company keep running 500 agents after one of them goes wrong, and Bhatia said rapid containment will be critical to protecting the country's progress as agentic AI becomes embedded across the economy.
Source: gulfnews.com
