In February 2026, Palantir tried to move a line the British government had drawn in public. The vehicle was not a tender. It was WhatsApp.
Messages obtained by the campaign group Foxglove under the Freedom of Information Act, and reported by the Guardian on 9 November 2026, show the company urging Zubir Ahmed, then a health minister, to consider applying its software to GP records. Those records sit expressly outside Palantir’s £330 million contract to run NHS England’s federated data platform, the system known as the FDP. The proposal has not been approved, and GP data is still outside the contract. What the messages show is a supplier, already inside the building, asking a minister to reopen a door that had been shut in writing.
That is a procurement story, not only a health story. Nothing released shows the ask was agreed. For anyone running workflow software so lodged in operations that leaving hurts, the risk transfers directly. Embedded vendors ask for the adjacent dataset. They ask in private. And the cost of walking away becomes the argument for staying.
The Department of Health and Social Care released the messages with the executive’s name redacted. The Guardian identified the sender as apparently Louis Mosley, Palantir’s UK boss in that report, and Palantir has not denied it. The redacted text does not prove the name, and the company’s ask does not depend on it. By the time the Guardian published, it described Ahmed as a former health minister. That is his status at publication, not a hint the texts went to someone already out of office. Britain appoints more than one health minister. He is not the same post-holder as Preet Kaur Gill, who answered questions on the award that June.
The logic they called unarguable, and the politics they knew were not
Palantir’s case, on its own terms, is not exotic. Join the data. Primary care is the GP surgery. Secondary care is the hospital and the specialist service. A frail patient’s story is split across both, and so are the waiting list, the discharge, and the chance to intervene before a crisis. The company told Ahmed that “the logic of joining up primary and secondary [care] data is unarguable, ” and that its platform, Foundry, was “ideal” for the job.
That operational case is not silly. A record split across a surgery and a hospital is one way people fall through a service. Calling the logic unarguable was the overreach. Palantir also told him the politics were not straightforward, because the previous government had ruled GP data out of scope when the FDP was announced. It urged him to “get out ahead of” the opposition it expected, and pressed for sponsorship rather than a seminar:
The FDP is already delivering very strong results, but will deliver even more benefits faster with the right ministerial sponsorship.
The benefits it attached to GP data were claims, not results: faster access to GPs, more effective and targeted management of frail patients, and more effective preventative care. NHS organisations, it said, would keep local control of their data. Rollout, Palantir has since said, is for government to decide. The Department of Health and Social Care declined to comment. The Guardian reports that Ahmed praised a BBC Sunday politics appearance shortly after Mosley had been on the programme. The released account shows the praise. It does not show him accepting or refusing the proposal. Warmth is not a decision.
The exclusion they were trying to soften is not a rumour. When Rishi Sunak’s government announced the deal, a parliamentary written statement on 21 November 2023 said: “GP data will not be part of the national platform.”
Trust is the constraint. Communications will not fix it.
GP records are a different class of file. They can hold a lifetime of identifiable clinical and personal detail: diagnoses, medications, mental health, family circumstances, the things a person tells a doctor and almost nobody else. Hospital operational data is sensitive. A lifelong primary-care record is the relationship itself, written down.
Doctors’ and patients’ groups, including the British Medical Association, have opposed bringing GP data into the FDP. Tom Dolphin, chair of the BMA’s council, has warned that access to identifiable GP data could undermine patient trust and the doctor-patient relationship. He has called the prospect scope creep: a supplier moving past the work it was hired to do. He put the worry this way:
Patients must be able to seek care without worrying which operators might have access to duplicates of their personal information.
That line is a warning, not a map of how access would work. The public record does not say who would be able to query identifiable GP data, whether a duplicate would be created, or how de-identification would work. Palantir’s answer is that NHS organisations would “maintain full local control” over their data. Local control is a promise, not a diagram. Federated systems are often sold as a way to analyse data across organisations without building one unrestricted central copy. Whether the FDP actually works that way is not set out in the public account of the contract.
Marc Farr, chair of the NHS Chief Data and Analytical Officers Network, has said the approach appears to stretch beyond assurances from ministers and NHS leaders, and that it would require legislative change. That is his assessment. The reporting does not identify the change he means, or a bill, a direction, or a ministerial decision.
Donald Campbell, Foxglove’s director of advocacy, said the messages showed Palantir pressing to bring GP records onto the platform despite 2023 pledges that this would not happen. He argued that the private channel was the point: the company knew how sensitive the ask was, and made it through privileged access rather than in public.
Palantir’s reply has been consistent. Data sharing, it says, can improve the NHS for patients and taxpayers. Its software helps NHS organisations do that while they maintain full local control. Government decides the rollout. The exchanges, the company says, reflect those points and nothing more ambitious.
There is a second trust problem, and it is not about waiting lists. The Guardian has reported Palantir contracts with the Israeli military, and contracts supplying AI for what it called Donald Trump’s homeland security crackdown. Those reports do not measure how an NHS scheduling tool performs. They do explain why a quiet expansion into the GP record was never going to land as a technical tidy-up.
Leaving looks expensive, which is how lock-in usually looks
Three days before the lobbying story, a different decision was being briefed, and the two should not be mashed into one headline. On 6 November 2026 the Guardian, citing the Financial Times, reported that the £330 million contract is to be extended. A review had found that triggering a break clause would be too disruptive. Ministers, the reporting said, plan a procurement for a UK alternative, with Palantir potentially replaced in about two years. The plan was expected to be signed off soon. That is a reported intention, not a signed extension, and it is not a decision to add GP records.
The disruption argument is not imaginary. The Commons Library, citing NHS England figures from August 2026, records that 142 of 202 acute, mental health, community and ambulance trusts (70 percent) were in live use, and 170, or 84 percent, had signed a memorandum of understanding. Thirty-five of 36 integrated care boards had signed and were live. That is deep embedding in secondary care and in system-level planning. It is not universal, and it does not touch GP records.
Use is not even mandatory. In June 2026, NHS England’s chief data officer, Ayub Bhayat, told the Health and Social Care Committee that NHS England had clarified the point.
An October 2025 Medium Term Planning Framework still says all NHS providers should be onboarded by 2028/29, using core products for waits, cancer, and urgent and emergency care, with integrated care boards using FDP tools for performance and population-health management.
A reported plan to line up a UK replacement within about two years of that November briefing sits next to an onboarding target that runs into 2028/29. Those are different documents. Read together, they describe a service still climbing onto a platform it is also, on paper, preparing to leave.
Critics, including Green party leader Zack Polanski, have called the evidence of patient benefit highly contested. The sequence matters more than the slogans. Government documents describe what the FDP is expected to support: elective recovery, vaccination, service planning, joined-up care, supply management, faster identification of patients on waiting lists, theatre and bed management, timelier discharge, less duplicated reporting. Those are aims. Palantir says the platform has helped cut surgery waiting times and discharge delays, and cites “very strong results.” The reporting attaches no number to either cut.
In February 2026 NHS England awarded Imperial College Projects a contract to evaluate the platform independently. The Commons Library says that evaluation is expected after the decision on whether to extend. The National Audit Office is examining how the Department of Health and Social Care and the NHS are integrating patient records, including through the FDP, and is expected to report in winter 2026/27. If ministers extend on the timetable being briefed, they will do it before a published independent evaluation exists.
Anyone who has renewed a major system because ripping it out would hurt operations will recognise the sequence. You do not need a conspiracy to get there. You need software already live in 142 of the 202 trusts NHS England counted, and an evaluation that has not yet landed.
Before the £330 million, a cheaper way in
The foothold predates the WhatsApp, and the public record describes it in layers that should not be collapsed into one price. During the pandemic Palantir provided software free, and then received paid contracts: £908, 000 in June 2020, £23.5 million in December 2020, and £11.5 million in August 2022. Separately, Louis Mosley (head of Palantir UK and Europe, in a Commons Library description, and the man the Guardian identified as the apparent sender of the later messages) told the Science, Innovation and Technology Committee in July 2025 that the software had been provided for a token £1, because the pandemic was a national emergency, not to buy an advantage. The £1 is his account of that emergency supply. It is not a fourth price for the three contracts, and it does not make those invoices unpaid.
Campaigners and some MPs have long argued that the emergency role advantaged the firm in later competition. In February 2023 the Government Chief Commercial Officer wrote to Palantir about the risk that low- or zero-cost footholds could undermine competition, a warning later reported by the National Audit Office. NHS England describes the FDP award as a competitive procurement and says any eligible supplier could apply. In June 2026, then health minister Preet Kaur Gill told the Health and Social Care Committee that officials named in press reports about Palantir links were not involved in awarding the contract.
Both points matter. A clean award process does not erase the structural question the commercial officer raised in 2023. Once a supplier’s software is already running beds, theatres and discharge across a large share of trusts, “any eligible supplier” lives more easily in a tender document than in a hospital on a Tuesday night.
Parliament has been sharper than a general murmur about trust. The Science, Innovation and Technology Committee’s June 2026 report, Rewiring the state, called Palantir’s healthcare role an “unacceptable point of weakness.” It warned of vendor lock-in, supplier dependency and weaker digital sovereignty from reliance on overseas providers. The committee said its findings were not a judgment on product quality or on ideology.
NHS England’s answer, as recorded by the Commons Library, is contractual and technical. Lock-in, it says, is limited by a maximum seven-year term committed only in increments, by exit planning, in-house skills, other partners and portable standards. Those accounts do not say whether the seven years run from award, signature or go-live, and a November 2023 announcement is not a start date you can multiply out. Incremental commitments, and the break clause described in the November 2026 briefings, are why an early exit was a live question at all, rather than a date at the far end of a maximum term. Those safeguards do not demonstrate a way to add, or refuse, GP data. They are the right list for a buyer who still has a pen.
An operational platform, whatever the label
Official descriptions of the FDP talk about an operational data platform. Guardian reporting calls it an AI system and talks about AI tools. Neither defines Foundry’s models, what is automated, or what a clinician would see if GP data were ever connected. The jobs officials have actually assigned (waiting lists, beds, discharge, performance) are the unglamorous core of what gets sold as AI for business. They are measurable. On GP records, there is nothing yet to measure, because the records are not in the system.
What is settled, and what is still a WhatsApp
-
Have GP records been added to Palantir’s platform?
No. The February 2026 approach has not been approved, and GP data remains expressly outside the £330 million FDP contract.
-
Did a minister agree to the idea in those messages?
Nothing released shows Zubir Ahmed accepting or refusing the proposal. Praise for a television appearance is not a policy decision, and the Department of Health and Social Care has declined to comment.
-
Is the contract definitely being extended?
Not in any document here. On 6 November 2026 the Guardian, citing the Financial Times, reported an intention to extend because triggering a break clause looked too disruptive, alongside a plan to procure a UK alternative and potentially replace Palantir in about two years. That plan was expected to be signed off soon. It is not a signed extension, and it is not a decision to add GP records.
-
Has the platform been shown to cut waiting times?
Not with a published number. Government papers list expected operational benefits, Palantir cites strong results, and critics call the evidence contested. NHS England commissioned Imperial College Projects in February 2026 to evaluate the platform independently. The Commons Library says that evaluation is expected after the decision on whether to extend.
-
Does “local control” answer the doctors?
Only as a claim. Palantir says NHS organisations would keep full local control. The public record does not say who could query identifiable GP data, whether a duplicate would be created, or how de-identification would work. The BMA’s worry is a warning, not a proven access model.
For buyers
Copy the unglamorous safeguards: incremental commitments, exit plans, portable standards, and an evaluation that lands before renewal. Refuse the pattern where a supplier reopens a written exclusion in private while the public story is still that the data is out of scope. That last line is a rule, not a finding. The findings are the five answers above.
The operational case for joining a GP record to a hospital record is serious. A lifelong identifiable clinical file is still not a customer record, and it is not a module you switch on because the vendor is already in the workflow.
The method is the problem. Palantir asked a minister, on a private channel, to stretch a boundary a previous government had stated in public. That decision has not been taken. If the NHS wants primary-care data analysed alongside hospital data, under this supplier or another, it can say so in the open, under whatever law that actually requires.
The warning for anyone with software already live in the workflow is smaller than the headlines. Your vendor will eventually ask for the adjacent dataset. Write the exclusion so a WhatsApp cannot move it.