Cops put Microsoft Copilot in holding cell after controversial hallucination
- Reference: 1769607607
- News link: https://www.theregister.co.uk/2026/01/28/microsoft_copilot_wmp/
- Source link:
The force's recommendation to block Maccabi Tel Aviv fans from attending the team's Europa League match against Aston Villa on November 6 was partly based on material generated by the AI tool about disruption to a non-existent match against London club West Ham.
Cop cops it after Copilot cops out: West Midlands police chief quits over AI hallucination [1]READ MORE
On January 6, former Chief Constable Craig Guildford told Parliament's Home Affairs Committee that his officers had not used AI to find this material, then corrected this in a letter to say they had.
Following criticism by Home Secretary Shabana Mahmood and others, Guildford [2]retired at age 52 . West Midlands Police and Crime Commissioner Simon Foster said that he had already started the process that would allow him to fire Guildford when he announced his retirement.
The force's acting Chief Constable, Scott Green, speaking at a public meeting with Foster on January 27, said he could not discuss the details of how Copilot was used as this was under investigation by the Independent Office for Police Conduct and West Midlands Police's professional standards department.
[3]
But Green [4]added : "On my first day in office, I took the immediate step to turn off access to Microsoft Copilot within the force. Like everybody else, we will be using AI in the future, we will use Microsoft Copilot, but we are not going to do so until I am satisfied with the conclusion of what has happened in this case and until we have got broader policy and guidelines in place."
[5]AI conference's papers contaminated by AI hallucinations
[6]AI superintelligence is a Silicon Valley fantasy, Ai2 researcher says
[7]AI music has finally beaten hat-act humans, but sounds nothing like victory
[8]OpenAI says models are programmed to make stuff up instead of admitting ignorance
Responding to a query from The Register , Foster said: "Artificial intelligence has an important role to play in policing, to increase productivity, improve crime detection, and manage emerging threats. However, use of AI must be ethical, lawful, transparent, and evidence-based, so as to maintain the trust and confidence of the public."
"Serious concerns have arisen about the use of Microsoft Copilot by West Midlands Police, when preparing to police the Aston Villa v Maccabi Tel Aviv football match in November 2025," Foster added. "It is important therefore that West Midlands Police pause its use, whilst resolving matters relating to the governance, oversight and regulation of its use, as a matter of urgency."
[9]
In the meeting, Foster also questioned Green about poor record keeping of meetings with Dutch police and findings of confirmation bias in the force's assessment of the risk from Maccabi Tel Aviv fans. Green apologized for the way the force had handled the work, describing the failings as "damning." He said the force is working to rebuild trust with Jewish communities by investigating every report of antisemitic hate crime and introducing antisemitism training for staff.
In a white paper [10]published this week , the government laid out plans for a big increase in the use of AI in policing, spending £115 million over the next three years on a new National Centre for AI in Policing to be known as Police.AI that will initially focus on automating administrative work. ®
Get our [11]Tech Resources
[1] https://www.theregister.com/2026/01/19/copper_chief_cops_it_after/
[2] https://www.theregister.com/2026/01/19/copper_chief_cops_it_after/
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aXpAuaCBdMEen3oeUohVFwAAAQE&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://westmidspcc.public-i.tv/core/portal/webcast_interactive/1049003/start_time/5400000?force_language_code=en_GB
[5] https://www.theregister.com/2026/01/22/neurips_papers_contaiminated_ai_hallucinations/
[6] https://www.theregister.com/2025/12/11/ai_superintelligence_fantasy/
[7] https://www.theregister.com/2025/11/17/ai_music/
[8] https://www.theregister.com/2025/09/17/openai_hallucinations_incentives/
[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aXpAuaCBdMEen3oeUohVFwAAAQE&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[10] https://www.theregister.com/2026/01/28/tech_in_policing_white_paper/
[11] https://whitepapers.theregister.com/
This isn't a CoPilot, AI or LMM issue. The problem is that someone simply believed (or confirmed what they wanted to believe) what the internet told them without checking sources or, even better, getting a quallified expert view.
"I know it's true, don't argue with me. I read that on the internet, it's everywhere!"
Too many people like this lately.
Yes and no.
In fact this is a very good example of AI hallucinations and bias - and how to fail by ignoring both.
So the root cause _is_ an LLM/AI/Copilot issue.
However, you are right in that the user should have checked/verified the LLM output.
Which show's today's AI problems in a nutshell:
- You get AI-generated answers/drafts incredibly fast.
- Verifying every bit of generated information will take much longer than getting the initial answer.
- It might even take longer than writing a less-detailed answer yourself
- You boss expects you to be "more efficient" with the expensive AI, you now command...
So guess, where corners will be cut?
Re: Yes and no.
I've had colleagues creating some user panic because of info they got from Google's AI summaries being delivered as "the truth". And those colleagues are technical people, they just seem to lack critical thinking....
These are the same type of people who get their lorries stuck in little villages because they blindly believe their GPS systems, even when the road is clearly labelled "not suitable for HGVs". The bigger problem is that they are allowed to get way with it, by retiring on a nice pension instead of being sacked.
Like everybody else, we will be using AI in the future
Marvellous: policing by fad.
The only thing worse than relying on AI is a policeman relying on AI
“I have never seen a situation so dismal that a policeman couldn't make it worse.”
― Brendan Behan
Re: The only thing worse than relying on AI is a policeman relying on AI
Although also in fairness: To err is human, but to really f**k things up you need a computer
A huge assumption ..
.. is being made by this article that AI was in anyway actually responsible for WMP's actions. WMP claim it was, but there is every reason to suppose they are throwing copilot under a bus to save their own skins.
WMP have repeatedly lied about so many diff aspects of this story, including lying and giving contradictory and misleading information to parliament a few weeks ago that it is entirely reasonable to doubt anything they say.
And I say WMP, but I really mean the small cadre of senior WMP coppers, especially the now RETIRED ON FULL PENSION (rather than sacked & prosecuted) former Chief Constable Craig Guildford, who took all the decisions that led to what happened. I suspect all the other coppers in WMP are as angry about events as everyone else.
"WMP have repeatedly lied about so many diff aspects of this story"
Correct, that is the problem. Guildford explicitly and emphatically denied about WMP's use of AI when coming to the decision they did when originally quizzed in the HoC.
"He could not discuss the details of how Copilot was used as this was under investigation by the Independent Office for Police Conduct and West Midlands Police's professional standards department."
I am absolutely certain he could discuss it, but, just like his predecessor, it was doubtless yet another deep hole of inconvenient truths that the proles shouldn't need to worry themselves about.
I suspect the use of AI was tangential. Stuff I've read about this suggests, they simply wanted a report to justify a position they'd already taken. Because "community groups" would object to the fans being allowed to come to the match. So in order to have an easy life, they had to come out with reasons to ban them. Hence there's a dispute about what Dutch police said to them in a meeting that they appear to have "lost the minutes of", and the report with the fake AI results.
AI makes a terrible search tool, because if you ask it to find stuff it will try to find that stuff, and if there isn't any/enough, it'll helpfully invent it for you. Whereas it's much better at summarising text you give it, which I guess makes it a better research tool.
Seeing as the AI tools seem to have taken on some of the hidden assumptions of the language, I also wonder if you might need to phrase your prompts carefully so that you do not bias the results. Ask it to find examples of something, I suspect it's more likely to do so - even if it has to invent those examples.
may as well police by facebook rumor
and just think, millions of people are now taking search result AI generated overviews as gospel.
It’s the ultimate lazy human solution.
Those of us who have the inclination to at least verify facts by researching have always been in the minority.
Re: “always in the minority”
Hence “The Minority Report”.
Can you give me a cite ?
ChatGPT: Sure - here's a Copilot link.
Copilot: Sure - here's a Claude link
Claude - Sure - here's a Grok link
That's odd. I'm sure a report I read at the time that use of Copilot was banned in the force. Maybe Iwas hallucinating if it's only being banned now.