Defence AI readiness is a data problem, not a model problem
In January 2025 the Defence Select Committee's Second Report of Session 2024-25, Developing AI capacity and expertise in UK defence (HC 590), documented a rhetoric-reality gap in MOD's use of AI. The fragmentation it named, separate AI activity across each frontline command, will not be closed by better models. It will be closed by asset and operational data the Department can actually trust.
What Parliament actually said
HC 590 found that while MOD policy describes AI as transformative, the Department was not behaving as though that was the case. Expert oral evidence was blunter. Neil Morphett, Chief Engineer at Lockheed Martin Rotary and Mission Systems UK, was asked whether the authority was AI-ready. His answer: "the answer is probably no." The Committee named fragmentation across frontline commands as a specific barrier and flagged SQEP recruitment into Defence AI roles as insufficient.
That finding sits alongside JSP 936 v1.1, Dependable Artificial Intelligence (AI) in Defence (Part 1: Directive), issued 13 November 2024. It is the principal policy framework governing the safe and responsible adoption of AI in MOD. The policy is in place. The practice is catching up.
MOD is moving, but not on the data layer
A serious reader will know MOD has stood up the Defence AI Centre, published the Defence AI Playbook, and is investing in cross-command coordination. None of that is in dispute. What those initiatives do not do, and were not designed to do, is reconcile the underlying asset and operational record across the Defence estate, its single-service teams and PFI partners, OEM sustainment systems and logistic information systems. The data layer is the layer the JSP 936 reviewer will ask about. It is the layer this piece is about.
DSIT and the Government Digital Service published the Digital and Data Benefits Framework in April 2026, sitting alongside HM Treasury's Green Book as the standard methodology for quantifying the benefits of digital and data programmes. Any Defence AI business case that goes through a Green Book review is now scored against the DSIT methodology. The data section is where most Defence programmes have never been asked to articulate realised value; the service transformation and interoperability sections are where the weakest proposals will lose marks twice.
Fragmentation is a governance problem, not a technology one
The same tail number, the same barracks block, the same generator set can appear differently in three different records, with different dates and different custodians. A model trained on that record will confidently reproduce the ambiguity. JSP 936's ethical assurance requirements (data provenance, traceability, human oversight) cannot be satisfied by a supplier that does not know where its training data came from.
The SRO reads this as an assurance problem. The Programme Director reads it as a data-reconciliation problem. The Commercial Lead reads it as a supplier-clearance problem. A single reconciled, governed, SC-cleared asset and data record is the artefact that answers all three.
Three areas where the data record decides the outcome
Estate and basing decisions. Consolidation, resilience and net-zero pressures converge on one question: what is the current condition and utilisation of the estate, and what does that imply for the next ten years? Answering with AI requires a reconciled record across the Defence estate, its single-service teams and PFI partners. Most programmes start by discovering how little of that reconciliation has been done.
Platform sustainment. Availability-based contracts, predictive maintenance and AI-led scheduling only deliver if the asset record is complete, classified consistently and governed. A model that confidently predicts the wrong failure mode, on the wrong platform, in the wrong squadron, is worse than no model at all.
AI assurance for the supply chain. JSP 936 places ethical assurance on the Department and its suppliers. Most suppliers have never had to evidence data provenance to a review board. Fewer still have done so under SC or Secret conditions.
Why SC clearance matters at the data layer
Big 4 consultancies bring brand and scale. They also bring global delivery models in which the data work (labelling, cleansing, integration, enrichment, quality assurance) is routinely offshored. For Defence AI, that is precisely the layer that cannot leave the boundary. Offshored data work breaks the JSP 936 provenance chain before the model is even built. The data work itself, the discovery, the reconciliation, the governance artefacts, has to be performed on-shore by cleared practitioners. That is increasingly specified at RFQ level as the Department tightens supplier expectations.
ISO 55000 is the right starting point
AI governance frameworks often start from the model outwards. That works for consumer AI. It does not work for a platform or an estate whose asset record is the primary input. ISO 55000 starts from the asset outwards, and that difference of starting point is material. Correctly applied, it gives an auditable structure that a JSP 936 review board and a Permanent Under Secretary can both sign off. The mapping from the standard onto a Defence AI programme is specialist work; it is where most programmes would benefit from outside help.
The continuous layer: Data Governance as a Service
Defence data does not become defensible once and stay defensible. Platforms change, suppliers rotate, estates consolidate, sustainment data ages. JSP 936 assurance is not a single event either. The assurance cycle is continuous, and the record behind it needs to be.
Data Governance as a Service (DGaaS) is Brainwave Asset Intelligence's cross-sector model for that continuous layer: SC-cleared, on-shore governance that detects duplicates, degradation and supply-chain integration gaps in the asset and operational record, between assurance cycles rather than at them. The full treatment, applied consistently across asset-intensive sectors, sits in the DGaaS anchor on this page.
Foundations before automation
The Defence AI conversation needs to start lower down the stack than it currently does. Before the model, before the capability, before the tooling, the question is whether the asset and operational record underneath is complete, governed and defensible. Parliament has said it is not. The doctrine (JSP 936) now requires that it be. The supply chain is working out what that means.
Data readiness is the foundation. It is also where Brainwave Asset Intelligence sits: SC-cleared, ISO 55000-disciplined, on-shore, platform-agnostic across Maximo, Octave (formerly Hexagon), Planon, Ultimo, Infor and SAP EAM. Foundations before automation.
Key takeaways
- The Defence AI readiness gap is a governance and data problem, not a model problem. HC 590 documented the rhetoric-reality gap; expert evidence to the inquiry gave a straight "probably no" on AI-readiness.
- JSP 936 v1.1 is mandatory and flows to the supply chain. Most suppliers are still building what it requires.
- The Defence AI Centre and AI Playbook do not substitute for a reconciled asset record. That layer is still to be done.
- SC-cleared on-shore delivery is a sector prerequisite. Offshore data work breaks the provenance chain.
- ISO 55000 gives Defence AI readiness the auditable structure a JSP 936 review board will sign off.
Sources: House of Commons Defence Committee, HC 590, Developing AI capacity and expertise in UK defence, 2nd Report of Session 2024-25, 10 January 2025 (witness: Neil Morphett, Chief Engineer, Lockheed Martin Rotary and Mission Systems UK). Ministry of Defence, JSP 936 v1.1, Dependable Artificial Intelligence (AI) in Defence (Part 1: Directive), 13 November 2024. DSIT and Government Digital Service, Digital and Data Benefits Framework, April 2026. HM Treasury Green Book. Verified 20 April 2026.