Technology

Add as a preferred source on Google

UN Guide Calls for Human Oversight of AI in City Services

The 167-page ITU and UN-Habitat toolbox ranks uses by potential impact and cites Rotterdam’s decision to stop a welfare risk model after an audit found bias risk.

Updated 2 min read
Delegates listen during a smart cities policy session in Barcelona, Spain. (AI-generated image / EconoTimes)
Delegates listen during a smart cities policy session in Barcelona, Spain. (AI-generated image / EconoTimes)

A new United Nations guide asks cities to keep identifiable officials accountable for AI-assisted public services, making human control and risk-based safeguards central to decisions affecting residents.

The International Telecommunication Union (ITU) and UN-Habitat introduced the 167-page United for Smart Sustainable Cities (U4SSC) toolbox in Barcelona on Nov. 4. It advises municipal authorities to define the public problem first and compare AI with simpler or non-digital ways of addressing it.

That choice matters for residents whose access to services could be affected by automated systems. The guide says cities can forgo AI when alternatives pose fewer risks, datasets fail to reflect people affected, or authorities lack the capacity to govern a system. Anacláudia Rossbach, UN-Habitat executive director, said meaningful public participation requires “addressing gaps in data, protecting human rights and ensuring that residents and civil society meaningfully participate in shaping urban decisions.”

The toolbox groups AI uses into lower, medium and high impact. Decisions affecting housing, benefits or social care, as well as critical infrastructure, emergency dispatch and public-space surveillance or biometric identification, are among its high-impact examples. For those uses, it recommends impact assessments, human oversight, independent review or audits, and testing before deployment.

The guide cites both efficiency gains and reasons to withdraw a system. In eThekwini, South Africa, an AI mapping tool generated more than 1.5 million building footprints in about 72 hours; comparable mapping work can take months or years. In Rotterdam, the city stopped using a welfare risk model after an audit identified a risk of bias and returned to random selection, the guide says.

The toolbox presents its measures as guidance rather than universal legal duties. Binding requirements depend on national, regional or supranational law. Where laws are silent, cities can set controls through internal policies, procurement conditions or contract clauses.

The U4SSC working group says pilot cities are to put the guidance into practice in 2027. It is inviting cities to register interest by March 31, 2027.

Related articles

View all