UAE firms lead the world in deploying AI agents, but most can't contain a rogue one quickly, survey finds

Companies in the UAE are running more AI agents in production than their counterparts in any other surveyed market, but only a small fraction could quickly shut down one that misbehaves, according to new research from software firm Dataiku.

The study, conducted by Harris Poll across eight markets, found that 62 percent of UAE chief information officers believe their organisation has more than 50 AI agents actively running in production. Fifteen percent said they operate more than 500, the highest share of any surveyed country and well above the 9 percent global average.

The scale of deployment does not appear to have brought correspondingly greater damage. Eighty percent of UAE CIOs said they had encountered an agent that violated business intent, policy or expectations while still operating within its technical parameters. However, 62 percent said the incident was caught before it caused customer, financial, compliance or operational impact. That was the highest proportion globally, compared with 48 percent overall. Just 18 percent reported an incident that resulted in actual impact, the lowest figure in the study and well below the 31 percent global result.

Detecting a problem and stopping it across an organisation are different matters, the research suggests. Only 5 percent of UAE CIOs said they could reliably identify and contain a problematic agent across environments within one to two hours, among the lowest results in the study and half the worldwide figure of 10 percent. Another 27 percent could act later the same day, meaning more than two-thirds could not reliably contain an agent on the day it was found. For 45 percent, containment would take 24 to 48 hours, and 23 percent said it would take two days or longer, one of the highest figures in the study.

The research points to fragmented oversight as a factor. Only 18 percent of UAE CIOs said they have a fully unified view of agents across platforms, data environments and business applications, and just 12 percent said agent lifecycle management is standardised across their organisation.

"CIOs in the UAE are running agents at a scale few markets have reached, and they're unusually good at catching bad behaviour before it does damage," said Florian Douetteau, CEO and co-founder of Dataiku. "But detection and control are different capabilities, and the research shows a real gap between them. An agent you can spot in one system but can't stop everywhere else isn't contained — it's just been noticed."

UAE organisations are also building up staff to support the expansion. Among CIOs whose companies are creating or planning embedded roles for AI experts or forward engineers, 22 percent expect to recruit more than ten people over the next year, the highest share among the countries studied.

Many are also preparing for disruption to the underlying technology. Nearly two-thirds (63 percent ) plan to use more or different models to strengthen business and AI continuity, while 62 percent are considering open-source alternatives as a hedge against future unavailability. The release said such measures could help organisations build AI sovereignty and insulate themselves against events such as the temporary suspension of Anthropic's Fable 5 and Mythos 5 models under a US government order.

Technology leaders are under mounting pressure to show results. All UAE CIOs surveyed (100 percent) reported greater board pressure to demonstrate measurable AI return on investment than in 2025, and 93 percent said their CEO has indicated their job security depends on AI outcomes. Ninety percent said they would stake their job on delivering measurable AI results.

Sid Bhatia, Dataiku's Area Vice President and General Manager for the Middle East, Turkey and Africa, said the UAE has built a strong foundation for agentic AI. "The next stage is to connect those safeguards across the enterprise," he said, calling for a unified view of agent populations, 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. Rapid containment, he said, will be critical to protecting the country's progress as agentic AI becomes embedded across the economy.