A one-page technology assessment before any public body buys an AI system

The problem: public bodies buy AI tools quickly, and the questions that matter, such as what it is for, who could be harmed, how errors are caught and how to stop using it, are often asked late or not at all.

The proposal: before any significant purchase, the buyer completes and publishes a one-page assessment. It states the purpose in plain language, the people affected, the data used, the main risks and how they will be tested, the person responsible for oversight, and how the system can be switched off or appealed against. Technology assessment is the practice of studying the likely effects of a technology on people and society before it is widely adopted.

Who acts: procurement offices require it; an independent group reviews a sample each year.

Costs and trade-offs: it takes a few hours of work. The risk is checklist thinking, so assessments should be written in ordinary language and reviewed by someone outside the buying team.

First step: pilot it for the next three AI-related tenders in one department.