AI & Technology

Trust Is Contextual. What Public Services Get Wrong About Citizen Consent

By Kevin Kimber, UK Senior Vice President and General Manager at Oracle

Government is deploying conversational and agentic AI into services citizens cannot opt out of. The citizen has no alternative provider, so trust must be built into the service itself. You can choose another bank after an automated failure; you cannot choose another tax authority or benefits system. That absence of an exit changes what trust means in the public sector, and it is why the citizen-facing AI debate cannot be borrowed wholesale from the private sector.

The industry often frames the challenge as perception: does the public trust AI, and how do we reassure people? But trust is not a fixed attribute that citizens grant or withhold. It is contextual. It moves with what is at stake, what is on offer and what recourse exists if something goes wrong. Someone who guards their personal data carefully in ordinary life may hand over everything in a medical emergency if sharing it could save them. Nothing about the person has changed. Only the stakes have.

The useful question is therefore not the broad ‘does the public trust AI?’ It is far more specific: under what circumstances, for what benefit and with what recourse would this citizen consent to this interaction if they genuinely understood it? That is a higher bar than a reassuring interface, and it is the one public services should build to.

Where Trust in AI Actually Sits Today

For the UK public sector, sovereignty is no longer simply a procurement checkbox. It is an architectural choice about where data is held, who is permitted to administer an environment and how isolated a service needs to be. The right model will vary with what is at stake: from globally distributed operations, through data held in-country and administered by cleared nationals, to complete isolation.

The principal underneath is simple: leave the data where it is and bring the learning to it. Many people assume that for AI to use data, the data must travel to an external system. It need not always. That assumption reveals how much anxiety about AI is really anxiety about information leaving the building. Sovereignty should be chosen service by service, matched to the sensitivity of the interaction, rather than treated as a constraint to apologise for.

The Interaction You Cannot Walk Away From

When a commercial AI interaction fails, the market can correct it: the customer leaves. When a public sector interaction fails, the citizen may absorb it, often when they are least able to absorb anything. The failure to design against is not merely inaccuracy. It is indignity: being processed by a system that got it wrong, in a way nobody can explain, with no visible route back to a human being.

A subtler risk sits behind the interaction. As services are automated, the human capability supporting them can quietly erode. A system that summarises a citizen’s case decides what matters and what does not. When it decides wrongly, the omission does not announce itself. Responsible automation must therefore keep the human capable, not just present.

Done well, technology gives people back the time and insight to serve residents better. The London Borough of Lambeth, after standardising processes on Oracle Cloud, has described using analytics to understand service delivery and free staff to support residents. That is the test for citizen-facing AI: not whether it automates an interaction, but whether it leaves the people behind it more able to serve the person in front of them.

Public bodies that get this right will be deliberate about three things: consent explained and given with a complete understanding, sovereignty chosen for each service, and a dignified route back to a capable human. Beneath all three is a dependency worth stating plainly. A trustworthy interaction cannot be built on an untrustworthy decision. Citizen-facing AI is only as dependable as the data governance and judgement beneath it. Trusted services rest on a trusted foundation.

The prize is not a better chatbot. It is an interaction a citizen would agreed to participate in, had they fully understood it. That is a far higher bar than most public sector AI is held to. It is also the only one worth clearing.

Related Articles

Back to top button