Britain's Police Just Drew a Line: No AI in Court Statements

The most common phrase on office desks right now might be: “Can’t we just get AI to do this?” Report drafts, emails, meeting notes — hand them all to a chatbot. But what happens when that “this” is a court statement that could send someone to prison? Police in England and Wales just stopped cold in front of that question.

On June 6, a story climbed Hacker News to 122 points and 41 comments. The headline was plain: police in England and Wales told to stop using AI for court statements. It reads like a dull bit of administrative housekeeping. I’d argue it’s the first clear line a justice system has drawn around AI.

What actually happened

The substance is simple. Officers across England and Wales were instructed not to use AI tools when drafting statements or evidence documents headed for court. And “AI tools” here means the everyday chatbots — specifically commercial services like Microsoft’s Copilot.

The reasoning surfaced in a comment from a senior policing figure, who said he had intervened with frontline units that were deploying commercial AI tools “before proper evaluation.” Translation: officers were already using unvetted AI in the field, and leadership scrambled to hit the brakes.

The telling part is what he added next — that every force “has a good policy” on Copilot use. Policies existed. Control did not. That gap between the policy document and what’s actually happening on the ground exists in every organization, and it’s on full display here.

Why court statements, specifically

Using AI to tidy up meeting notes and using it to write a court statement are not the same activity. The thing that separates them is AI hallucination.

Generative AI is not a machine that looks up facts and reports them. It’s a machine that predicts the next statistically plausible word and strings sentences together. Which means it will invent things that never happened and deliver them in crisp, confident prose. We call that hallucination.

A hallucination in your meeting notes? You shrug, think “that’s odd,” and move on. A hallucination in a court statement is a different beast. An event that never occurred gets recorded as fact. Words nobody said get attributed to someone. A single document decides who’s guilty and who walks. There’s no undo button.

We’ve already seen this play out elsewhere. Lawyers in multiple jurisdictions have been publicly humiliated for submitting AI-fabricated case law — citing rulings that don’t exist, complete with invented case numbers. There’s no reason a police statement would be immune.

If the policy existed, why the ban?

Here’s the wrinkle worth sitting with. The official said good policies were already in place. So why issue a blanket stop?

My read: AI tools are too easy, too fast, too convenient. If writing a statement drops from 30 minutes to three, an overworked officer is not going to find it easy to resist that. A line in a policy document saying “use with caution” was never going to be enough.

Commercial AI tools carry a second problem on top. The information police handle is a dense pile of victim identities, investigation details, and sensitive personal data. The moment that gets typed into an external AI service, controlling where the data goes becomes nearly impossible. “Before proper evaluation” almost certainly carries this data-security worry inside it too.

How the community read it

Hacker News split in two. One camp called it obvious. Introducing unvetted AI into a domain where accuracy is everything — the justice system — is dangerous on its face.

One comment landed somewhere drier: surprise that AI would turn out to be the fork in the road toward “Idiocracy.” (For the unfamiliar: the film about a future where everyone’s gotten dramatically dumber.) The joke carries a real fear — that offloading human judgment to AI eventually costs us the judgment itself.

I’ll be honest about the data, though. This topic hasn’t been widely chewed over yet. Over the past 30 days there wasn’t much related discussion — a single Hacker News thread and a YouTube video on UK justice and AI are about the extent of it. This is less a roaring debate than a quiet but meaningful signal.

The takeaway: watch where the line falls

This guidance matters not because it blocks AI outright. It matters because a society, for the first time, gave a concrete answer to the question of how far to trust AI. Convenience is fine — but we won’t let AI near decisions that can’t be reversed. That’s exactly where Britain’s justice system drew its line.

And this isn’t only about police. Medical diagnosis, hiring decisions, loan approvals — every domain that shapes a person’s life is standing in front of the same question. Where is AI being used in your workplace? And which of those jobs is the kind where one hallucination, even once, is unacceptable? Drawing that boundary in advance might be the most necessary work there is right now.