London police in talks with Palantir over AI for criminal investigations

London police are discussing with Palantir the possible purchase of AI technology for criminal investigations, and this is not just another experiment with fashionable software. The systems in question could automate intelligence analysis and speed up work with information in cases where arrests, the testing of hypotheses and decisions on how an investigation proceeds are at stake. No contract has been signed yet, but the talks themselves show that Britain's largest police force is treating AI as a tool for everyday operational work, not merely as a supporting technology for pilot projects.

According to available information, Palantir gave a demonstration of its solutions in March 2026 to senior officers of the London police intelligence unit. Inside the force, staff responsible for analytics were tasked with finding processes that could be automated with AI to raise productivity. That detail matters: the police are not looking for abstract innovation but for specific areas where algorithms can sort through large volumes of information faster, cross-reference signals from different sources and cut manual work. For services that process reports, briefings, internal notes and data on active cases every day, even partial automation could noticeably change the pace of work.

Palantir is a natural fit for such a role. The company has long built its business on analytical software for governments, intelligence agencies, militaries and law enforcement. Its products are associated not with public chatbots but with systems that merge scattered data, help reveal connections and support decision-making in sensitive operational environments.

But that is exactly what makes a potential deal controversial. Palantir already has high-profile and politically toxic cases: its software is used by the US immigration service ICE as part of Donald Trump's hardline immigration policy, and by the Israeli military. For some staff and observers, this means the question is not only about effectiveness but also about the supplier's reputation, the company's political ties and the acceptable limits of cooperation between police and a private contractor.

Data adds its own tension. If London police do decide to deploy the Palantir platform, the company would potentially have to work with highly sensitive information: criminal investigation materials, intelligence summaries, and details about suspects, witnesses, sources and working hypotheses. Even if the contractor formally does not get full control over such datasets, the architecture of access, integration and data processing inevitably becomes a subject of dispute.

In systems like these, the key question is not whether AI can save time, but who exactly sees the data, where it is processed, how conclusions are verified and who is responsible for an error. For police this is especially sensitive, because any inaccurate link, false priority or opaque recommendation can affect the fate of a real person.

The interest in such tools is easy to understand. British law enforcement, like many public bodies, is trying to raise productivity without a proportional increase in headcount, while the volume of data in investigations keeps growing. In this context AI promises not to replace the detective but to lift part of the repetitive analytical load: speed up sorting of information, highlight important matches and help prioritize lines of inquiry. But in criminal justice the cost of error is higher than in ordinary corporate analytics. So every such deployment quickly goes beyond an IT purchase and becomes a public debate about rights, accountability and the limits of automation in law enforcement.

What this means. The talks between London police and Palantir show that AI is moving ever deeper into the core of police analytics, where people and closed internal processes used to dominate. If the deal goes ahead, it will be a strong signal to other European agencies: commercial AI platforms are starting to be treated as investigative infrastructure, not as a peripheral experiment. But the story also shows something else: processing speed is no longer the only criterion. The origin of the technology, control over sensitive information and the state's ability to explain to the public where useful automation ends and risk to civil liberties begins are just as important.