TL;DR
A new study built on policy analysis, surveys of residents and interviews with practitioners concludes that councils are deploying AI faster than the governance needed to make that deployment responsible. The researchers set out six priorities, and find public trust — not technical performance — consistently shaped how people judged the risks.
What councils are already running
The deployments run further ahead than the debate suggests. Two UK councils now use a chatbot called ELLIS, able to field questions across a thousand-odd areas of council business. Cumbria County Council brought in a tool that processes certain parking permits and school invoices without staff involvement, cutting more than a tonne of paper in a quarter. A London borough has used AI to reduce by as much as 60% the hours social workers spend drafting case notes and assessments, keeping consent, privacy protections and human review in place. Elsewhere, a planning body has trialled automatic summaries of public submissions.
The pattern repeats internationally: Amarillo in Texas fields around 16,000 resident enquiries monthly through a virtual assistant, avoiding a call centre expansion costing US$1.8 million; Boston trimmed average junction delays by 13.5% with AI-optimised traffic signals; Helsingborg fitted refuse vehicles with image recognition to spot potholes.
The six priorities
The researchers name clear accountability — defining where AI may be used, who answers for it, and at what point a human must be involved; organisational capability across elected members, executives and staff; data and technical safeguards; transparency plus a route to human review; genuine community participation; and workforce preparation so the technology augments staff rather than displacing them.
Their framing matters more than the list. Responsible AI is usually treated as a broad policy aspiration; for councils it is hardening into a concrete public expectation that varies from place to place. Across the five jurisdictions studied — including Australia, Spain and Hong Kong — trust consistently shaped how risk was perceived.
Looking forward
For UK readers this lands beside two live domestic stories. Angus Council recently signed a £240,000 contract for an AI note-taker for social workers, and North Lanarkshire has just admitted passing a resident’s details to a data centre developer after a planning enquiry. One is a procurement decision, the other a data handling failure, and neither is a model performance problem — they are precisely the accountability and safeguard gaps this research describes.
Councils stay answerable for outcomes even when a system makes the call. Better algorithms will not close that gap; visible oversight, published limits and a working route to human review might.