The government wants AI agents throughout public services, but it is keeping decisions about people’s livelihoods in human hands.
The United Arab Emirates plans to expand artificial intelligence across government while withholding one category of authority from the technology: an AI agent will not be allowed to make a decision affecting a person’s livelihood. Waheed Abbas of Khaleej Times reported Wednesday that UAE Minister of State for Artificial Intelligence Omar Sultan Al Olama said governments still do not know enough about the full capabilities of these systems to give them that level of control.
The restriction is arriving inside one of the world’s more aggressive government automation efforts. Federal employees are being trained to use AI, and the UAE is preparing to deploy autonomous agents across public services rather than limiting the technology to chatbots and clerical assistance. Al Olama said half of federal employees have completed extensive AI training, with the remainder expected to finish by next August.
Abu Dhabi is already reporting substantial productivity gains. Haneen Dajani of Khaleej Times reported that AI allows government employees to redirect roughly 280,000 working hours a year away from routine processing. Teams using the TAMM government-services platform can launch and quality-check services 70% to 80% faster, according to the Department of Government Enablement.
Those gains create pressure to automate more of the decision chain. Once software can gather records, compare eligibility criteria, apply rules and recommend an outcome, allowing it to complete the final decision can look like a small additional step. It removes another layer of labor and increases the amount of work one system can process.
The UAE is stopping short at the point where administrative efficiency becomes public authority.
A government employee who denies a benefit, rejects an application or approves a license operates inside a chain of responsibility. There is an agency, a supervisor, a policy and usually some process for review. AI agents make that chain harder to see. A consequential decision may depend on a model developed by one company, trained on data collected somewhere else, configured by another team and deployed by an agency whose staff may not be able to explain every inference the system made.
Automation can therefore move decision-making power away from identifiable public employees without creating an equally clear place for accountability to land.
Keeping a human responsible for livelihood decisions prevents that transfer from becoming complete. AI can still prepare the file, compare information, detect anomalies and recommend an outcome. A public official remains responsible for deciding whether the recommendation should become government action.
The policy also acknowledges a problem that accuracy testing alone cannot resolve. A system could eventually become statistically better than a human at evaluating certain applications and still raise questions about whether the government should delegate that authority. Public decisions involve discretion, explanation, appeal rights and responsibility in addition to raw predictive performance.
Al Olama has applied similar reasoning to more extreme uses of autonomous systems, arguing that countries should pursue international agreements around AI weapons capable of deciding to kill. The difficulty, as he acknowledged, is that governments hesitate to limit themselves when they believe competitors may refuse to do the same.
Livelihood decisions provide a nearer-term version of the same governance problem. Public agencies are adopting AI now, and the pressure to hand increasingly complex work to autonomous systems will grow as the technology improves and budget savings become easier to quantify.
The next regulatory fight may therefore center less on whether AI is technically capable of making a consequential decision and more on which forms of authority governments are willing to delegate. Countries establishing limits now are deciding whether public power remains attached to accountable human officials or gradually migrates to systems whose decisions may be efficient, fast and difficult to challenge.