Five British police forces have decided to step away from Palantir’s Foundry system after two years of testing it on criminal investigations, and they want proof the company has erased what it held.
The East Midlands Special Operations Unit, which ran the pilot, concluded it would not buy the platform, the BBC reported on October 11. The unit also asked the U.S. company to demonstrate that it has severed its connections to police systems and deleted the related data.
The trial began in October 2024 and fed data from Leicestershire Police into the AI and machine learning platform, which was used in investigations of serious and organized crime.
The reason given was practical rather than philosophical. The project lead said that with Home Office funding exhausted, the assessment of what had been delivered and the arrangements for further money led the unit not to proceed.
EMSOU is a regional body that pools specialist resources for several forces across the East Midlands, the kind of buyer that technology vendors court precisely because a single contract can serve more than one department.
Palantir built its business in government, and police work has been one of its more sensitive frontiers. Foundry ingests data from many sources and presents it in a form investigators can search, which is exactly the capability that makes police forces interested and civil liberties groups uneasy.
The decision is the ordinary end of a pilot that did not convince its sponsor. Read against the way enterprise software is sold to the public sector, though, it carries more signal than a single non-renewal usually does, because pilots are where vendors live or die and renewals are the only measurement that matters.
Two parts of the outcome deserve attention. The first is the funding. Home Office money carried the trial, and when it stopped, the case for continuing had to stand on its own. It did not, at least not yet, which suggests that the value the platform delivered was not visible enough to survive a budget review.
The second is the deletion demand. A police force asking a departing vendor to prove it has cut its links and removed its copies of case data is treating the exit as an exercise in evidence, not trust. That is the correct instinct, and it is also a reminder of how much data a platform accumulates while it is being evaluated.
Verifying deletion is harder than requesting it. A vendor can confirm a contract has ended and that production access has been revoked, but proving that no copy survives in a backup, a development environment or a model trained on the data is a different exercise. EMSOU’s request puts the burden on Palantir to show its work.
For Palantir, the loss is small in revenue terms and awkward in precedent. The company has spent years arguing that its systems give public agencies capabilities they cannot build themselves. A pilot that ends without a purchase, in a country where it has pursued government work aggressively, is a data point its competitors will use.
Police forces across Britain have been weighing how far to allow algorithms into investigations. The technology promises to find links across databases too large for a person to read, and it raises the question of who is accountable when a connection is wrong. Serious and organized crime cases, with their long paper trails and multiple forces, are the setting where the argument is most acute.
The politics around police data have grown sharper in the past two years. Forces have faced questions about what they retain, how long they keep it and who can query it, and vendors have been drawn into those arguments because their platforms are where the data is combined. A system that organizes investigative material from several forces is useful precisely for the reason it attracts scrutiny: it puts information in one place.
Other forces have taken a different view of the same technology, which is why the outcome at EMSOU should be read as a decision about this pilot rather than a verdict on the category. Vendors in this market typically absorb several quiet exits for every contract they win, and the losses rarely produce a public statement.
The pilot also shows how the procurement pipeline works in practice. A two-year trial funded by a central department, run by a regional unit, evaluated against a budget cycle it did not control, will end when the money does unless someone with a permanent budget decides the result is worth keeping.
That is the gap between a successful demonstration and a contract, and it is where most police technology quietly stops. The capabilities can be real, the officers can find the tool useful, and the program can still close because the next fiscal year has no line for it.
EMSOU has not published the assessment that produced the decision, so the specifics remain with the unit and the company. What is public is the sequence: a trial that began with Home Office support, a review after the support ended, and a conclusion not to buy.
For Palantir, the question now is whether this is an isolated budget casualty or a sign that British police buyers, having tested algorithmic investigation, are less willing than their American counterparts to convert the tests into long contracts. The next round of renewals at other forces will answer it more clearly than this one did.
