Will replacing Palantir solve the NHS’s fragmented governance problem?
Replacing a supplier can change who operates a system, but it does not remove the complexity surrounding the system.
In February 2027, the government initiates the termination clause for Palantir's services, paving the way for a new vendor to enter the scene at a more favorable cost and with a British postcode. The migration process commences. The federated data platform, operational across various levels, maintains responsibility and control in the hands of data controllers.
However, the program's projected lifetime cost of £1.1 billion versus projected benefits of £808 million has sparked scrutiny over its business case. The Department of Health and Social Care has acknowledged the need to review the contract before the break clause expires, with questions surrounding value, sovereignty, and supplier dependency.
The Federation Data Platform (FDP) was specifically designed as a federated system, with trusts and integrated care boards managing local instances while NHS England oversees the national instance. This design fosters multiple layers of ownership, which can enhance accountability by keeping control within the relevant data controllers.
However, this federated model also generates technical dependencies across systems and organizations that must be continually mapped, risk-assessed, and managed. The complexity of the existing estate, with three-quarters of providers identifying investments needed to address technical debt, presents ongoing challenges.
Despite the temptation to replace Palantir with a British supplier, the underlying governance issues persist. The data controllers, legacy systems, and technical dependencies remain unchanged. The NHS possesses numerous governance roles and frameworks, such as the Data Security and Protection Toolkit, Cyber Assessment Framework, data protection impact assessments, senior information risk owners, Caldicott Guardians, and Digital Technology Assessment Criteria.
Yet, these measures primarily assume that an organization can assess a system at a specific point in time, assuming the resulting picture will remain reliable. In the era of AI-native data federation, this assumption becomes increasingly untenable.
AI systems and agents connecting to existing applications, datasets, and third parties expand the governance surface beyond the data platform itself. Security teams must now possess visibility into more than just the underlying infrastructure, understanding the ownership of each AI system and its permitted reach and actions. This dynamic is not a future concern; by the end of May 2026, 139 trusts and 35 integrated care boards were already live on the FDP, creating a substantial network of local and national instances, not a single platform or supplier.
Existing AI assurance approaches are insufficient for this level of dynamism. The Medicines and Healthcare products Regulatory Agency's (MHRA) approach to AI as a medical device and NHS England's AI governance guidance provide valuable foundations. However, periodic assessments are inadequate to govern an agent whose access or behavior can change at runtime.
The risk extends beyond a poor AI recommendation; it includes the inability to quickly establish what the system could access, what has changed, or who is accountable for the outcome.
When evaluating the next platform, the focus should shift from cost or sovereignty to continuous visibility and accountability across the broader estate. Practical requirements include identifying AI systems operating across the estate and maintaining a usable inventory of models, data sources, owners, and approvals. The system should reassess risk when models, data, or permissions change, rather than relying on annual returns or scheduled reviews.
Controls should align with the specific use case, recognizing that a discharge-scheduling tool and a cancer-pathway tool carry different risk levels. Real-time evidence and audit trails should be generated, turning compliance into a query rather than a fire drill. In the event of an incident, accountability must persist, identifying the legal controller under UK GDPR when a trust's data passes through a national instance, determining who owns incidents that start locally and end nationally, and identifying who approved particular access grants, models, or actions.
These answers should be recorded within the organization's operating system, connected to the decisions and permissions they describe, rather than reconstructed from documents written months earlier.
The issue of governance extends beyond the NHS, as other sectors, including water, energy, transport, and local government, are also constructing connected estates with layered ownership and suppliers they may not fully understand. The break clause may alter the supplier for the NHS but will not solve the governance challenges. A true solution must enable the NHS to identify what data runs across the estate, which systems can access it, who is accountable for data protection controls at each level, and what emerging risks arise as the system continues to evolve. Without this capability, changing the contract name is merely a distraction.
Rewritten by urgent.news; may contain errors — check the original.