60,000 Opt-Outs in Two Months: UK Health Minister Warns Palantir 'Mistrust' Is Now a Research Problem

For three years, the debate over Palantir’s role in Britain’s National Health Service ran along familiar lines: privacy campaigners against procurement officials, anecdote against dashboard. On September 7, 2026, that debate crossed a threshold. A serving UK government minister went on the record to say that public “mistrust” of the US defence and health tech company has become an operational risk to the NHS itself — because patients are responding with their feet, or rather with their records.

James Frith, the health innovation minister, made the admission in a letter to Layla Moran, the Liberal Democrat chair of the Commons health and social care committee. His concern, as reported by The Guardian: “mistrust” of Palantir and “the impact it could have on people’s willingness to share data with the NHS.”

The number that changed the conversation

Behind the ministerial language sits a hard figure. Between mid-May and mid-July 2026, roughly 60,000 additional people registered a national data opt-out — the mechanism that prevents an individual’s medical records from being used for research and planning beyond their direct care. Frith described the rise as “modest,” but conceded it tracks closely with media interest in Palantir’s position, and said it would continue to be monitored.

Sixty thousand opt-outs in sixty days is not a privacy story in the abstract. It is a measurable subtraction from the datasets that underpin British medical research. The UK’s research advantage has always rested on a cradle-to-grave health system covering an entire population, with (historically) unusually high public willingness to allow secondary use of records. Every opt-out erodes the representativeness of that asset. Frith put the stakes bluntly: “It may not be possible to realise the benefits of the 10-year health plan if patients stop sharing their data.”

There is a bitter irony here that the minister did not spell out. The federated data platform (FDP) was sold as the engine of that 10-year health plan. If trust in the platform’s operator causes patients to withdraw the very data the platform exists to process, the contract begins to defeat its own purpose — a self-inflicted wound priced at £330 million over seven years.

A quiet U-turn on mandates

The letter’s second bombshell is a policy reversal delivered almost in passing. In July 2026, a health minister told Parliament that NHS trusts would be told they “must use” parts of the federated data platform, which is controlled using Palantir technology. Frith now writes that “there is no requirement to use the FDP” and that trusts may “prefer an alternative.”

For a programme whose value proposition is integration — one platform, one picture of the health system — making adoption optional is more consequential than it sounds. Greater Manchester’s integrated care board has already demonstrated the exit path, remaining the only ICB in England to decline the Palantir-built platform. If adoption becomes a trust-level choice, the “federated” in federated data platform starts to look aspirational.

Moran welcomed the “softened stance” but went further: “We remain of the view that the government should switch provider and not extend the current contract. Given how long this process could take, ministers need to act decisively and quickly, and not put this decision off until February.”

February is the operative word. The contract contains a break clause, and doctors’ and patients’ groups are openly campaigning for Labour to exercise it, citing Palantir’s work with the Israeli military and the US Immigration and Customs Enforcement agency, and questioning the company’s value-for-money claims.

The contested scoreboard

Palantir’s defence rests on operational numbers, and they are not trivial. A company spokesperson said trusts using the platform have recorded “110,000 additional operations, a 15% reduction in discharge delays for long-stay patients and a 6.8% improvement in the number of people finding out whether or not they have cancer within 28 days.”

But the statistical ground under those claims is soft. NHS England has reported similar figures while explicitly saying it cannot “draw conclusions about cause and effect” — correlation, in other words, not attribution. Trusts that adopt a new data platform are also trusts that have just invested in improvement; separating the tool from the enthusiasm is genuinely hard. The UK government’s statistics watchdog, the UK Statistics Authority, is now investigating the data. A BMJ-published analysis in June found NHS England admitting the data don’t prove the contentious software benefits the health service.

A pattern of escalation

The opt-out surge did not appear from nowhere. 2026 has delivered a steady drumbeat of Palantir controversies:

• February: Health officials’ own briefings, revealed by The Guardian, feared Palantir’s reputation would hinder delivery of the “vital” contract.
• April: Alarm among health service staff after Palantir engineers were given NHS email accounts.
• May: MPs warned that NHS England had granted Palantir access to identifiable patient information for its AI plans — prompting ministerial apologies.
• July: The Commons health committee itself urged the government to ditch the contract; London mayor Sadiq Khan separately blocked the Metropolitan police from awarding Palantir a £50m AI contract, a decision the company is now suing over.
• August: NHS England’s data protection officer confirmed in writing — in response to a campaign group — that patients cannot opt out of the platform itself, only of secondary uses, a distinction that fuelled further public anger.

Tom Hegarty of Foxglove, the tech equity campaign that has led much of the scrutiny, said tens of thousands opting out “in just the last few months” vindicated the warnings: “People are right to be concerned given … its founder, Trump ally Peter Thiel, going on record to opine that the NHS makes people sick.”

The bigger lesson for AI in public services

Strip away the specifics and the Palantir affair is a case study in a failure mode that will recur wherever AI platforms touch public data. The procurement logic treated trust as a constant — an assumption baked into the business case, not a variable to be managed. It isn’t. Trust is an input that degrades when the operator’s other activities (military contracts, immigration enforcement) clash with the values a public service is expected to embody, and it degrades measurably, in opt-out registries.

The economics are unforgiving. A 15% improvement in discharge delays is worth a great deal. But if the platform’s association costs the research base even a few percent of population coverage — and 60,000 opt-outs in two months suggests the slope can be steep — the net ledger looks very different. Data-quality losses from differential opt-out are not evenly distributed either: they skew toward the people most distrustful of the state and of American tech companies, which is to say toward precisely the communities health research most needs to reach.

Frith’s letter marks the moment the UK government stopped treating public mistrust as a communications problem and started treating it as infrastructure risk. Whether that recognition arrives in time to save the research datasets — or the £330m contract itself — depends on a decision ministers are now under pressure not to defer past February.