ICO: Royal Free NHS Trust broke data-protection law sharing 1.6 million patient records with Google DeepMind for the Streams kidney-injury app

The problem

England’s NHS trusts hold identifiable records on millions of patients, and acute kidney injury is a fast-moving, often-missed condition that clinicians wanted earlier warning of. To build a detection tool, Royal Free London NHS Foundation Trust entered a data-sharing arrangement with Google DeepMind. The Trust provided the personal data of “around 1.6 million patients” as part of a trial to “test an alert, diagnosis and detection system for acute kidney injury” [source]. The open question the regulator later answered was whether patients had any real say in, or knowledge of, that transfer.

What was built

DeepMind built Streams, a mobile app to alert, diagnose and detect acute kidney injury, and to develop and test it the Trust transferred identifiable records in bulk. TechCrunch described the arrangement as sharing “1.6 million patients’ medical records” to power “an alert system for acute kidney injuries” [source]. The Information Commissioner’s Office opened an investigation into whether that transfer complied with the Data Protection Act 1998 ([source]).

The outcome

The regulator found the arrangement unlawful. “The ICO has ruled the Royal Free NHS Foundation Trust failed to comply with the Data Protection Act when it provided patient details to Google DeepMind” [source]. Independent reporting the same day carried the same verdict: TechCrunch wrote that the Trust “failed to comply with the Data Protection Law” ([source]), and The Register put the transfer at “1.6 million patient details” [source].

The basis of the finding was transparency and patient expectation. Information Commissioner Elizabeth Denham said patients “would not have reasonably expected their information to have been used in this way, and the Trust could and should have been far more transparent with patients as to what was happening” [source]. She framed the wider stakes as a trade-off that should not be forced: “the price of innovation does not need to be the erosion of fundamental privacy rights” ([source]).

Rather than a fine, the regulator required remediation. The Trust was asked to “establish a proper legal basis under the Data Protection Act for the Google DeepMind project and for any future trials”, to set out how it would comply with its duty of confidence in future trials, to complete “a privacy impact assessment, including specific steps to ensure transparency”, and to “commission an audit of the trial, the results of which will be shared with the Information Commissioner” [source]. Denham said the ICO had “asked the trust to commit to making changes that will address those shortcomings, and their cooperation is welcome” ([source]).

Weakest load-bearing source, named. The strongest evidence here is the ICO’s own published determination, a Tier 1 regulator source, but the exact artifact quoted is the ICO press release retrieved through a wired-gov mirror of the official release, not the full undertaking letter the Trust signed [source]. The independent press restatements from TechCrunch and The Register corroborate the finding and the 1.6-million figure, but they report the regulator’s conclusion rather than making an independent legal determination of their own ([source]). The finding itself is unambiguous and on the public record; what would lift the ceiling is the primary undertaking and the later independent audit quoted directly.

How this was verified. Method (checked 2026-09-10): the headline facts (around 1.6 million patient records, the “failed to comply with the Data Protection Act” ruling, the acute-kidney- injury purpose, the Denham quotes and the four required changes) were quote-matched verbatim against the Information Commissioner’s Office finding of 2017-07-03, retrieved via the wired-gov mirror of the official ICO press release and saved to sources/ico-press-release-2017-07-03.txt - Tier 1, a regulator’s own published determination. The ruling and the 1.6-million figure were independently corroborated the same day by TechCrunch (sources/techcrunch-2017-07-03.txt) and The Register (sources/theregister-2017-07-03.txt) - Tier 2, independent press. The quoted ICO primary is the press release, not the full undertaking/letter PDF, which is the main path-to-green lever. web.archive.org save and availability endpoints were unreachable from the war-room box this session, so an archive gap is flagged and the local snapshots stand in. No confirmation was sought from Royal Free or DeepMind: asking the subject of a finding to confirm it is a testimonial, not an audit.

Related case files

• Babylon Health’s AI-first primary care collapse — another heavily hyped UK health-AI venture built around the NHS whose story ended badly, here in the market rather than with a regulator.
• ICO orders Serco Leisure to halt unlawful facial-recognition and fingerprint attendance monitoring — the same regulator, the ICO, finding another organisation processed personal data without a lawful basis and ordering it stopped, the closest governance analogue to this ruling.
• Meta paused training its AI on EU public data after the Irish DPC asked it to delay — a data-protection regulator intervening on personal data fed to AI, the same “was there a lawful basis and did people expect this” question, seven years on and at far greater scale.