CENTRALISED SYSTEMS

On 12-13 November 2026, a fault inside one company's servers took email and Teams away from NHS trusts, GP surgeries, ambulance services and police forces across the UK. It was not a cyber attack. Nobody broke in. Microsoft's own systems stopped trusting their own servers, and the failure spread through a single shared infrastructure to 74 countries in about an hour. Clinical care largely continued because the NHS still has an older, distributed layer underneath its digital one: telephones, text messages, paper records and people on site. That layer is the real resilience in the system, and it is not being designed for, funded or protected. The same week, ministers were reported to be preparing a two-year extension of Palantir's NHS Federated Data Platform (FDP). The FDP is described as federated because each trust holds its own instance. But every instance runs the same proprietary software, from one supplier, under one contract and one legal jurisdiction, with one AI orchestration layer routing data to third-party models. It is distributed in hardware and centralised in control. Its failure domain is the vendor, not the trust. This paper argues that resilience is a property of architecture, not of vendors or nationality.