London’s mayor has blocked a proposed £50m Metropolitan Police deal with Palantir, according to the Guardian. The argument is partly about procurement rules, partly about public trust, and partly about a much bigger question: how should public bodies use AI when the stakes are serious?
The reported deal would have seen the Met use Palantir’s AI technology to automate intelligence analysis in criminal investigations. The Mayor’s Office for Policing and Crime, which has to approve large contracts, vetoed the plan after raising concerns about how the procurement had been handled.
For ordinary readers, this is not just a London politics story. It is a useful case study in why AI used by councils, police forces, the NHS and other public bodies needs more than a promise that the software is powerful. It needs proper competition, clear oversight, privacy protections, value-for-money checks and a way for the public to understand what is being done in their name.
What the row is about
According to the Guardian, the mayor’s office said there had been a “clear and serious breach” of procurement rules and that the Met had only seriously considered one supplier: Palantir. The Met criticised the block as disappointing and argued that it needs modern technology to work faster against organised criminals and hostile states.
Both sides are pointing at real issues. Police forces do handle huge amounts of information, and better tools may help investigators spot patterns, connect cases and reduce wasted manual work. At the same time, policing data is highly sensitive. A system that analyses intelligence can touch people’s privacy, reputations and rights, even when they have not been charged with anything.
That is why the process matters. Public-sector AI is not the same as downloading a new app. If a public body becomes dependent on one private supplier, changes the way sensitive data is processed, or spends tens of millions of pounds, the checks around the decision are part of the technology.
Why supplier choice matters
One concern in the Guardian’s reporting is “vendor lock-in”. That means an organisation becomes so dependent on one supplier’s system, data formats, workflows or support that switching later becomes expensive or impractical.
This is a familiar technology problem, but it matters more in public services. If a police force, hospital trust or council builds a core workflow around one AI platform, the supplier can become hard to replace even if costs rise, performance disappoints, or public confidence changes.
That does not automatically mean a particular supplier should be banned. It means the public body should be able to show why it chose that supplier, what alternatives were considered, what the exit plan is, who owns the data, and how the system will be audited.
ManyHands recently covered why AI tools need careful checks before being given access. The same principle applies here, but at a much larger scale. The more access an AI system gets, the more important the safeguards become.
AI can help, but it should not hide responsibility
A common mistake with public-sector AI is to talk as if the software itself is the decision-maker. In reality, people choose the supplier, set the rules, decide which data goes in, interpret the outputs and act on them.
If an AI tool flags a person, a pattern or a case as important, someone still needs to understand why. Someone needs to decide whether the output is reliable enough to use. Someone needs to be accountable if the system is biased, wrong, poorly configured, or used beyond its original purpose.
That accountability can get blurry when systems are complex or proprietary. If officials cannot explain how a tool is being used, what data it handles, how errors are caught, and what people can do if something goes wrong, public trust will suffer.
What ordinary people should watch for
Most of us will not be reading procurement documents for police technology contracts over breakfast, which is probably a mercy. But there are useful questions to ask whenever a public body announces a new AI system.
What problem is it meant to solve? What data will it use? Was there open competition? Who can audit it? Are humans still making the important decisions? Can people challenge mistakes? Is there a clear limit on how the data and system can be used later?
Those questions are not anti-technology. They are the boring but necessary plumbing that makes serious technology usable in public life. Without them, AI can become a black box that is trusted because it sounds modern rather than because it has been properly tested.
The bigger lesson for UK AI
The government and public bodies are under pressure to use AI more quickly. That pressure is understandable. Many services are stretched, budgets are tight, and staff are drowning in admin and information overload.
But speed is not the only test. A system used in policing, healthcare, benefits, tax or education can affect people directly. If it goes wrong, the consequences are not just a bad recommendation or an annoying chatbot answer. They can involve missed services, unfair treatment, intrusive surveillance or decisions that are hard to challenge.
The London row is a useful reminder that AI adoption is not just about whether the tool works in a demo. It is about the terms on which society accepts it. Public bodies need to be able to say: here is why we need this, here is why this supplier was chosen, here is what the system can and cannot do, and here is who is responsible if it fails.
That may sound less exciting than “AI will transform policing”. It is also the bit that decides whether the public can trust the transformation.
Source: The Guardian.
