London's mayor has blocked the Met Police's proposed £50 million Palantir deal, but the sharpest lesson is not only about politics or ethics. It is about procurement discipline, and why public sector AI vendors can no longer treat oversight as a late-stage formality.
The Mayor's Office for Policing and Crime has refused approval for a two-year Metropolitan Police contract worth up to £50 million with Palantir, the US data analytics company co-founded by Peter Thiel. The decision stops what would have been Palantir's largest British policing deal and turns a local procurement fight into a wider warning for AI companies selling into government.
As The Guardian reported, Deputy Mayor Kaya Comer-Schwartz wrote to Met Commissioner Mark Rowley saying the force had seriously engaged with only one potential supplier, failed to obtain prior approval for its procurement strategy, and had not demonstrated value for money. She described the failure as a 'clear and serious breach' of applicable procedural requirements. That is unusually direct language for a contract dispute, and it matters because it shifts the argument away from abstract concern about AI and into the harder ground of process, cost, and accountability.
The proposed contract would have used Palantir's technology to automate intelligence analysis for criminal investigations. The company already has more than £600 million in UK public sector contracts, including work connected to the NHS, the Ministry of Defence, the Financial Conduct Authority, and smaller police forces. A recent Met trial involving Palantir technology to monitor staff behaviour had been carried out under a contract valued just below the threshold requiring City Hall approval. The larger criminal intelligence deal crossed that line.
Procurement Rules Became The Real Constraint
The decision rests on a procedural point, but it is not a small one. Met investments above £500,000 require MOPAC approval, a rule meant to provide independent scrutiny before large public commitments are made. City Hall said the Met had been reminded of that requirement, yet still moved ahead in a way that left officials unable to assess whether the market had been properly tested.
That is where the story becomes bigger than Palantir. The blocked deal shows how procurement rules can become a practical limit on police AI, especially when systems are expensive, proprietary, and difficult to unwind once embedded. If a force cannot show it tested alternatives, challenged pricing, and planned for future exit, the technology may never get to the deployment stage.
The value question also sharpened the dispute. Comer-Schwartz noted that the contract had originally been estimated at £15 million to £25 million a year, with the proposed deal landing at the top of that range. She was not satisfied that the Met could fund it without unacceptable pressure on other budgets. For a police force already facing operational strain, that is not just an accounting issue. It is a question of what gets displaced when a single technology contract absorbs tens of millions of pounds.
For vendors, the lesson is direct. Public sector buyers are being pushed to prove that AI contracts are competitive, explainable, and resilient. Sole-supplier engagement may be faster in the short term, but it creates legal, reputational, and political risk. In this case, that risk was enough to stop the deal.
The Ethics Argument Has Not Gone Away
Khan's office also raised a broader point that procurement law does not easily resolve. A City Hall spokesperson said there remains a question over whether a company's values and ethics should be considered when public money is awarded. Palantir has faced sustained criticism over its work with US Immigration and Customs Enforcement and the Israeli military, and campaigners have pushed UK public bodies to cut ties with the company.
That part of the debate is unlikely to disappear. Public services increasingly depend on private data platforms, yet voters have limited visibility into how those relationships are formed. When the supplier is politically controversial and the system touches policing, health, defence, or financial regulation, the contract is no longer just a technology purchase. It becomes a test of public trust.
There is also the risk of vendor lock-in. The College of Policing's AI procurement guidance warns forces to think carefully about supplier dependence, including how relationships end and whether systems can work without permanent reliance on one company. City Hall said the Met risked becoming locked into Palantir's technology. That concern is familiar from other public sector software deals, where switching costs rise once data models, workflows, and staff habits are built around a single platform.
What Happens Next
The block does not permanently exclude Palantir. MOPAC said it wants to work with the Met on a new procurement at pace, and there appears to be no formal bar on Palantir bidding again. That makes this a reset rather than a ban. The company could still compete if the Met runs a process that tests the market properly and secures approval at the right stage.
For AI founders and larger vendors, that distinction is important. Government customers still want tools that help process evidence, speed investigations, and make stretched teams more effective. Bedfordshire police, for example, has credited Palantir technology with helping officers process large volumes of evidence in an organised crime case. The demand is real.
But demand is no longer enough. Public sector AI contracts now need transparent pricing, credible alternatives, open standards where possible, and a clear exit plan. Those are not optional extras for later negotiation. They are part of whether a deal can survive scrutiny in the first place.
The Met can come back with a cleaner process, and Palantir can come back with a compliant bid. What has changed is the balance of power. Oversight bodies now have a fresh example of how to stop a major AI contract before it becomes inevitable. Any vendor selling proprietary systems into policing should read that signal carefully.
