Government AI Plans Risk Repeating Past Failures, Experts Warn
The United Arab Emirates announced plans this month to run half of its government services through autonomous AI systems within two years. The systems would operate as "executive partners" that analyze, decide and execute without human intervention.
Experts say the plan is reckless and could set a dangerous precedent for governments worldwide.
History shows the risks
The Netherlands provides a cautionary case. A self-learning system wrongly accused roughly 35,000 families of childcare benefit fraud between 2015 and 2021. Parents were ordered to repay tens of thousands of euros they never owed. More than 2,000 children entered state care. Homes were lost.
The discrimination was built into the system's design. Dual nationality and foreign-sounding names were flagged as risk factors. The scandal ultimately led to the resignation of Dutch Prime Minister Mark Rutte.
Australia's experience was similar. The Robodebt scheme pursued 433,000 welfare recipients for A$1.7 billion in supposedly unlawful debts between 2015 and 2019. Mothers testified that their sons killed themselves after receiving debt notices they could not challenge. A Royal Commission later found the program "neither fair nor legal."
In the US, Arkansas and Idaho replaced nurses with algorithms to assess home care eligibility. People with cerebral palsy, quadriplegia and multiple sclerosis saw their care cut by 20 to 50 percent overnight. Courts eventually halted the systems, but some patients were left without adequate support, leading to preventable medical complications.
Three core problems emerge at scale
When a caseworker makes a mistake, one person suffers. When an AI system fails, thousands can be affected before anyone notices.
The second issue is opacity. Agentic systems make decisions in sequence, with each step building on the last. By the time harm becomes visible, the causal trail is lost. Arkansas's algorithmic health system was so opaque that even its creators could not fully explain how it worked. A federal court described it as "wildly irrational."
Trade secrets and proprietary frameworks can hide this lack of transparency further.
The third problem inverts the burden of proof. Citizens must prove their innocence rather than requiring the state to justify its actions. Those least able to fight back-people with limited time, money, language skills and legal access-suffer most.
Accountability disappears
The UAE claims its program is guided by "people come first." The design suggests otherwise. A government that measures success by speed of AI adoption is tracking the wrong metric.
Speed is what vendors care about. A government's responsibility is duty of care grounded in human judgment.
Citizens expect government to be accountable and transparent. They expect explanations for decisions that affect their rights. When governments embrace autonomous decisionmaking for efficiency, they sign away that accountability.
Every algorithm scandal in recent years raises the same question: Who is in charge? In a government run by agentic AI, that question has no clear answer. The system decides, updates itself and moves on. Citizens have no recourse.
The path forward
Democratic accountability erodes not through open power grabs, but through procurement decisions that quietly displace human oversight. This undermines trust in institutions at a time when it is already fragile.
The UAE has the resources and talent to build a different model-one that augments human decisionmaking rather than replacing it. Other governments watching this experiment should demand the same.
The costs of failure would not be confined to the UAE. A single mother in another country could lose benefits cut by an algorithm she never knew existed. Countless others like her would bear the weight.
Learn more about AI for Government and AI Agents & Automation to understand how these systems work and what safeguards matter.
Your membership also unlocks:
