NHS removes list of 137 Palantir trusts: 29 of them don't use the platform
NHS England published a list of 137 hospital trusts that were supposedly already benefiting from Palantir's federated data platform. The Guardian surveyed the trusts: 29 said the system is not live at their site or that they do not use any of its applications. The list was then quietly deleted, replaced by a general page with no figures. The £330m contract is, as it happens, being considered for extension to 2029.
AI-processed from Guardian; edited by Hamidun News
A classic enterprise AI story: the vendor and the customer paint a pretty picture of deployment, and then it turns out that "deployed" and "in use" are two different things. Only here the scale is national, and the stakes are a £330m contract.
What happened
NHS England published on its website a list of 137 hospital trusts that are "live" and "benefiting" from the Palantir-based Federated Data Platform (FDP). The platform is sold as a tool for three tasks: improving patient outcomes, making better use of operating theatres, and discharging people faster to free up beds.
The Guardian contacted these trusts directly. 29 of them replied that the platform is either not live at their site or that they do not use any of its applications. According to the newspaper, some others use only a fraction of the system. After the story was published, the detailed list disappeared from the site. In its place is a short page saying that early adopters are "already seeing real results": less paperwork, faster patient pathways, more time for care.
Some of the data was gathered by the advocacy group Medact, which sent freedom of information requests to all hospitals and integrated care boards in England. NHS England did not respond to requests for comment.
Not the first time
The NHS has been through something similar before. The statistics regulator (OSR) previously warned that the body had no evidence for its claim that the technology helped carry out more than 100,000 additional operations. The OSR is now separately examining how the NHS describes the scale of the system's use.
Critics, including Labour MP Ben Coleman, point to a systemic problem. Claims are published without verification and then undermine trust in the platform itself. In his words, such "overclaiming" could put off other hospitals.
A political fight is under way in parallel. Ministers are reportedly considering extending the contract to 2029. Palantir has hired Labour peer Tom Watson as a lobbyist. The Unison union and a number of medical and civil rights organisations are demanding the contract be ended. They point out that it contains a "transition assistance period" that would allow a move to other solutions to be completed as early as 2027. Palantir responds that dropping it would amount to "politics over patients". Coleman, for his part, prefers the idea of an open data platform on which both the trusts themselves and British tech companies could build tools.
Takeaways for practitioners
No personal opinion on Palantir is required here. The more important lesson is about how to measure the adoption of AI and analytics.
- Separate "connected" from "in use." A "customer is live" metric says nothing about value. Count active users, the number of working use cases, and the share of processes that actually run through the system.
- Verify impact figures before publishing them. A claim like "100,000 additional operations" without a methodology is a ready-made target for regulators, journalists, and competitors.
- Build an exit into the contract. Here, the opponents' key argument is the built-in transition period. If you are buying a data platform, specify data export, portability, and shutdown timelines up front.
- Think about data sovereignty. For healthcare, the public sector, and finance, the question of "who holds the data and how do we leave" can outweigh functionality.
Limitations
All details about the actual use of the platform rely on the responses of the hospitals to the Guardian and on Medact's requests. There is no complete independent picture across all 137 trusts, and NHS England has not set out its own version. The opposite has not been proven either: the claimed benefits may be real where the platform genuinely works. The dispute now is over how widely it is used and what exactly is backed by data.
The bottom line is simple: in AI and data projects, a nice customer list is no substitute for honest usage statistics. And the more expensive the contract, the more rigorously those statistics will be checked.
- Apr 27, 2026Palantir, Thales and Air Space Intelligence compete for FAA's AI air traffic contract
- Apr 30, 2026Google pulls AI search feature that surfaced medical advice from random users
- May 2, 2026Palantir to lose access to New York City hospital health data
- Nov 12, 2026Palantir pushes to add GP records to £330 million NHS platform
Key Questions Answered
What happened?
NHS England published on its website a list of 137 hospital trusts that are "live" and "benefiting" from the Palantir-based Federated Data Platform (FDP). The platform is sold as a tool for three tasks: improving patient outcomes, making better use of operating theatres, and discharging people faster to free up beds. The Guardian contacted these trusts directly. 29 of them replied that the platform is either not live at their site or that they do not use any of its applications.
What are the key details on Not the first time?
The NHS has been through something similar before. The statistics regulator (OSR) previously warned that the body had no evidence for its claim that the technology helped carry out more than 100,000 additional operations. The OSR is now separately examining how the NHS describes the scale of the system's use. Critics, including Labour MP Ben Coleman, point to a systemic problem. Claims are published without verification and then undermine trust in the platform itself.
What are the key details on Takeaways for practitioners?
No personal opinion on Palantir is required here. The more important lesson is about how to measure the adoption of AI and analytics. Separate "connected" from "in use." A "customer is live" metric says nothing about value. Count active users, the number of working use cases, and the share of processes that actually run through the system. Verify impact figures before publishing them.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.