Governments Are Using AI To Draft Legislation. What Could Possibly Go Wrong?

When British officials reviewed evidence for an independent water-sector overhaul last year, they ran into a familiar civil-service problem: tens of thousands of submissions, all needing review — fast — before the window for reform closed. So ministers turned to an in-house AI tool called Consult, part of the “Humphrey” suite, which the UK government said sorted more than 50,000 responses into themes in around two hours. It cost £240, they claimed, followed by 22 hours of expert checks — a workload that, scaled up across government, could save 75,000 days of manual analysis each year.

A UK government spokesperson tells Tech Policy Press that “AI has the potential to transform how government works — saving time on routine administrative tasks and freeing up civil servants to focus on what matters most: delivering better public services for the British people.” They added that they’re ensuring its responsible use across government with guidelines and audits.

“Government services are easily flooded with requests,” says Chris Schmitz, a researcher at the Hertie School in Berlin and a fellow at the Center for the Governance of AI. Having studied AI use across governments and politics, he argues the real challenge isn’t just analyzing consultation material, but preventing the process itself from being gamed.