Mistrust of Palantir is now showing up in the NHS opt-out numbers

The national data opt-out does not touch a patient's own care. It stops their confidential patient information being used for planning and research beyond that care, which is exactly the pool clinical researchers and public health teams draw from ^[9]. So the roughly 15,000 people a month who registered one over the past four months ^[21] were not walking away from a product they had used and disliked. They were removing themselves from a dataset, and by the department's own description each removal is cumulative and difficult to reverse ^[10]. The letter is dated 21 August, the committee published it on 2 September, and the Guardian reported its contents on the Sunday, drawing out what the rise means for medical research ^[2][22]. Teams that build on someone else's platform often treat the supplier as invisible to users, since people see the service and not the subcontractor behind it, but Frith's letter shows that assumption failing under mistrust of Palantir specifically. He attributes the recent rise to increased media interest rather than to anything the programme itself has done ^[6], which is a departmental way of saying that the variable moving the number is a name in a headline. The scale cuts both ways, as the letter concedes. Against England's population of roughly 57 million, 60,000 is about 0.1%, which is why the department can call it small ^[7]. Add the roughly 20,000 spike when the Federated Data Platform was announced in 2023 ^[5] and the running total is near 80,000, about 0.14% of the population ^[19]. The 2021 GP Data for Planning and Research scheme took the national opt-out rate from below 1% to above 5% ^[8], which on the same population is a swing of at least 2.28 million people, roughly 38 times the recent movement ^[20]. Frith offers that comparison as proportion; it also shows how far this kind of problem runs once it starts. Most of the letter answers a different question. The NHS owns the intellectual property in the data models and the products built on the platform and could migrate them, so it is not locked in ^[14]. Three Palantir engineers hold administrative-level access for back-office technical support and have no permission to use the data for their own purposes ^[15]. Imperial College London is running an independent evaluation, the National Audit Office is reviewing the programme, and the Office for Statistics Regulation found no defects serious enough to require NHS England to withdraw its published evidence on the platform's value ^[16]. Those answer questions about security, contract and evidence quality. None of them is the question a patient answers by filling in the opt-out form. On the calendar, the commercial and legal assessment of the 330mn pound deal completes in the autumn, and Frith warns that if it concludes a successor contract is needed, standard government procurement timelines make it unlikely one could be in place by March ^[17][13]. The committee asked in July for the February 2027 break clause to be used ^[11], and Moran said it was time to crack on with finding an alternative in time for spring 2027 ^[12]. For any team carrying a supplier its users can name, two things matter here: whether the distrust attaches to something built or something bought, and whether the remedy a user reaches for is reversible. Where the answers are "bought" and "one-way", the supplier has stopped being a procurement line and become part of the product, and the cost of keeping it accrues in a number no feature release rebuilds. The tradeoff sits in the same box: switching costs migration and, on the government's own account of procurement timelines, probably a gap ^[17], and nothing in the letter says opt-outs return when a supplier leaves.