When something goes wrong with a government AI system in Singapore, there is a document that explains who is responsible. The Infocomm Media Development Authority published the world's first comprehensive governance framework for agentic AI — AI systems that act autonomously on a user's behalf — in the public sector on January 22, 2026 Singapore IMDA. It covers four dimensions: risk bounding (assessing and limiting what can go wrong before deployment), human accountability with override mechanisms, technical controls including pre-deployment testing, and end-user responsibility Baker McKenzie. Compliance is voluntary, but organizations remain legally accountable for their agents' behavior Two Birds.
When something goes wrong with a government AI system in the United Arab Emirates, the accountability architecture is, by the cabinet's own announcement, beside the point.
On April 23, 2026, the UAE cabinet published its ambition: deploy agentic AI across 50 percent of government sectors, services, and operations within two years UAE Cabinet, a target Gulf News confirmed the same day. Sheikh Mohammed bin Rashid announced the target, with a taskforce chaired by Mohammad Al Gergawi and overseen by Sheikh Mansour bin Zayed. Ministers and directors-general will be assessed on adoption speed — not on whether the deployed systems function correctly UAE Cabinet.
The UAE has published no governance framework. It has published no equivalent to Singapore's four dimensions. It has not, in any public document the cabinet has released, described an oversight mechanism for agents acting in its name.
Mohamed Roushdy, CIO at Reem Finance, told Computer Weekly the 50 percent target is achievable if the work is AI-assisted rather than fully autonomous — but that full autonomy at scale faces a harder ceiling: trust, governance, and accountability gaps that cannot be bridged by infrastructure alone. A separate IDC analyst, Manish Ranjan, assessed the UAE's digital infrastructure as strong enough to support the deployment in principle, with the real determinant being whether the data and process layers underneath are agentic-ready Computer Weekly.
The UAE is not starting from zero. It has built some of the most sophisticated digital government infrastructure in the region over two decades and is using this initiative to raise the bar for the broader GCC market Computer Weekly. But moving from digital maturity to agentic readiness is a different kind of problem. The first requires building systems that work. The second requires building systems that work and that you can explain, override, and be held responsible for when they do not.
Singapore published its framework in January, three months before the UAE announced its target. The sequence matters. The country most often cited as the UAE's peer in digital government had already written the document the UAE's announcement conspicuously does not reference.
The cabinet has given itself two years and a 50 percent coverage target. The structural question the announcement does not answer is this: when a deployed agent makes a decision that harms a resident or a business, who steps forward.
That question is not rhetorical. It is the difference between a government that uses agents and a government that is accountable for them.
