-
Technology
-

What a Credit Score Chatbot Reveals About the NHS's Data Promises

By
Distilled Post Editorial Team

Experian's decision to let British users summon a live credit score inside ChatGPT rests on a single, load-bearing assurance: that the model doing the talking never actually sees the data doing the informing. Somewhere behind the conversational interface sits a retrieval layer, a permissioned pipe carrying a person's financial history into a chat window without, Experian insists, exposing it to the reasoning engine that generates the reply. It is a neat architectural trick, and a commercially convenient one, since it lets a credit bureau meet millions of younger consumers where they already are without appearing to hand a foundation model the keys to their financial life. The trouble is that the NHS has just run an almost identical experiment, made an almost identical promise, and had to publicly admit that the promise did not hold.

In May, NHS England confirmed something its own Data Protection Impact Assessment had explicitly denied: that engineers working for Palantir, the American contractor building the Federated Data Platform, could reach identifiable patient records inside a supposedly restricted environment called the National Data Integration Tenant. Three Palantir staff held administrative-level access; a further thirty-three contractors across various suppliers held narrower, project-specific permissions. None of this was disclosed in the document that exists precisely to disclose such things. The error surfaced only because the National Data Guardian, Nicola Byrne, pressed the programme team for clarification after months of public correspondence through a campaign called Not With My NHS Data. NHS England apologised, corrected the paperwork, and insisted the access was technically necessary and tightly governed. Byrne, for her part, said she was not in a position to independently verify that claim. Her sharper point was structural rather than personal: confidence, once built on an assurance that turns out to be wrong, does not simply reset when the paperwork is fixed.

That is the real reason a consumer credit tool matters to anyone running an NHS data programme. Experian is conducting, at far lower clinical stakes, exactly the kind of trust experiment the health service has spent three years failing to land cleanly. It is asking the public to accept that a sensitive personal record can be piped into a general-purpose AI product while remaining architecturally sealed off from the model that produces the answer. NHS England has made a version of that same claim about Palantir for years, describing the supplier as a mere processor with no meaningful reach into patient data. The FDP's own history now stands as the clearest available evidence of how such distinctions blur in practice, whatever the intention behind them.

The timing sharpens the stakes rather than softening them. NHS England is midway through a ten billion pound technology programme that embeds an AI triage tool directly into the NHS App and rolls out ambient voice technology to record and summarise clinical consultations, with a technical architecture document promised for September and full national coverage targeted for 2028. The MHRA's own commission into AI regulation in healthcare found that most respondents doubted the current framework was adequate on data governance and privacy, and most wanted stronger post-market surveillance. Health secretary Yvette Cooper inherits a programme where the public case for AI-mediated data access has already been dented once, at exactly the moment the health service is asking patients to extend that same trust across triage conversations, clinical transcripts and an eventual Single Patient Record.

None of this means Experian's claims are false, or that NHS England's assurances about ambient voice technology will prove similarly shaky. It indicates a change in the burden of proof. A data-isolation promise is no longer a technical footnote that regulators wave through on the strength of an architecture diagram. It is a public commitment that will be tested, and that the NHS's own recent history shows can fail quietly, through an unread paperwork error rather than a deliberate breach. For NHS leaders watching consumer AI push into ever more sensitive corners of ordinary life, the lesson from Experian's rollout is not that finance is following healthcare's lead. It is that finance is now the industry with less to lose if the same assurance turns out, once again, to have been stated more confidently than it could be guaranteed.