Defence Standard 05-138¶
Status: In flight (Plan 100). WI-0 (foundation work) currently under way.
Scope¶
Defence Standard 05-138 Issue 4 (July 2024) is the UK MoD cyber security standard that contractors must meet when delivering systems to the Ministry of Defence. Although 05-138 predates the AI-specific guidance, its control set is what defence-supply-chain customers are contractually held to, so the Secruna pack treats it as a first-class framework rather than a sub-set of the AI Playbook.
05-138 is a profile-based standard: every system is assigned a Cyber Risk Profile based on the sensitivity of the data it handles and the consequence of compromise, and the control set the system must meet flows from the profile.
Profiles¶
| Profile | Description |
|---|---|
| Very Low | Public information; minimal consequence of compromise. |
| Low | Internal information; limited consequence of compromise. |
| Moderate | Sensitive information; significant consequence of compromise. |
| High | Highly sensitive information; severe consequence of compromise. |
The profile is set at AI-system-creation time and drives which controls apply. Higher profiles inherit the controls of every lower profile.
Categories¶
Each control evaluates to one of three categories:
mandatory_control— the control is required at this profile level. Failure to meet it is a contractual breach.recommended_control— the control is best practice at this profile but not contractually required. Customers typically meet these for defence-in-depth.not_applicable— the control does not apply to this AI system at this profile.
Why it matters for AI¶
05-138 is not an AI standard, but every AI system in a defence context is also a software system, and software systems in a defence context fall under 05-138. The pack lets a defence-supplier customer maintain a single inventory of AI systems mapped against both the Defence AI Playbook (AI-specific obligations) and 05-138 (cyber security obligations) without double-entry.
Reference¶
UK Ministry of Defence, Defence Standard 05-138 Issue 4 (July 2024).
Plan references¶
- Plan 100 WI-0 — Foundation: profile model on
ai_systems, control taxonomy, 05-138 framework registration. In flight. - Plan 100 WI-1+ — Per-profile control YAML pack and matchers against the Plan 96 WI-3 category taxonomy. Scheduled after WI-0.
What's not in this pack¶
The cyber security controls 05-138 specifies are evaluated against
the same artifact_metadata surface as the AI rule books. Where
05-138 controls require evidence outside Secruna's reach (e.g.
physical security of a facility) the verdict is not_applicable_to_secruna
and the control is flagged for the customer to evidence externally.
Analysis¶
Scope¶
Defence Standard 05-138 Issue 4 (July 2024) is the UK MoD's cyber security standard for contractors delivering systems into defence. It applies to every supplier — prime, sub, integrator, or service-provider — whose contract references the standard, which in practice means most MoD contracts touching networked systems. Geographic scope is UK MoD operational reach but propagates through the supply chain to non-UK suppliers via contract clauses. In scope: any system handling MoD information assets, including AI systems whose training data, model weights, or inference outputs classify as MoD information. The standard is profile-based: each system is assigned a Cyber Risk Profile (Very Low / Low / Moderate / High) based on data sensitivity and consequence of compromise, and the applicable control set follows from the profile. Out of scope: systems handling no MoD information (internal supplier R&D, public-domain reference systems); operator physical security; non-cyber operational security. 05-138 complements but does not replace the Defence AI Playbook: AI systems in defence fall under both, with the Playbook handling AI-specific obligations and 05-138 handling the cyber overlay.
Key obligations¶
- Profile assignment — every system must be assigned a Cyber Risk Profile at creation, justified by data classification and consequence-of-compromise analysis.
- Profile-driven control adherence — the controls applicable to the assigned profile must be evidenced; higher profiles inherit lower-profile controls.
- Mandatory controls — a subset of controls per profile are contractually required; failure is a breach.
- Recommended controls — best-practice controls that are not contractually required but expected for defence-in-depth.
- Supply chain propagation — primes must flow 05-138 obligations down to sub-contractors handling MoD information.
- Evidence retention — control evidence must be retained and producible on request through the contract lifetime.
- Change management — material changes to the system or its data flows trigger profile re-assessment and control re-evaluation.
Our coverage approach¶
The pack adds a cyber_risk_profile column on the ai_systems
inventory entity (Plan 100 WI-0 in flight) so every AI system
carries its profile alongside its framework subscriptions. The
control taxonomy uses three categories (mandatory_control,
recommended_control, not_applicable) consistent with the Plan
96 WI-3 cross-framework category enum work. Controls are evaluated
against the same artifact_metadata surface the AI rule books use,
so an AI system inventory item gets a single verdict view spanning
05-138 cyber controls and Defence AI Playbook AI obligations
without double-entry. Mandatory controls map to verdicts that block
deployment until cleared; recommended controls produce informational
verdicts for defence-in-depth. The change-management obligation is
served by the platform's audit log: any artifact change that affects
profile-relevant attributes triggers re-evaluation of the
applicable control set. Supply-chain propagation reuses the
mechanism from the Defence AI Playbook pack — supplier-origin
artifacts surface contract-clause guidance to the customer.
Gaps¶
Per-profile control YAML pack (Plan 100 WI-1) is scheduled
after WI-0 lands; until then the control set is scaffolded but not
fired against customer artifacts. Physical-security controls
(facility access, secure-area handling) are explicitly outside
Secruna's reach — verdicts return not_applicable_to_secruna and
flag the control for the customer to evidence externally.
Cryptographic-control evidence — 05-138 specifies key-management
and crypto requirements that we partially evidence via connector
artifacts but cannot fully evidence without integration with the
customer's HSM or KMS; partial coverage is documented in the verdict
copy. Profile-recalculation tooling — when a customer changes a
system's data classification, the platform re-evaluates the control
set, but the customer-facing migration UX (how to walk through the
new gaps) is on the Plan 100 WI-2+ backlog. Counsel-reviewed
contract clauses for supply-chain propagation are pending the UK
defence-law specialist counsel hire.
Customer impact¶
05-138 enforcement is contractual rather than regulatory. The practical consequences for a non-compliant supplier are sharp. Bid disqualification: MoD invitations to tender increasingly require evidence of 05-138 capability before the technical evaluation; inability to evidence the standard fails the bid gate. Contract breach: an existing supplier failing a 05-138 audit faces remedies that range from corrective-action plans through fee withholding to contract termination, depending on control criticality. Supply-chain reach-back: a sub-contractor failure becomes the prime's problem — primes have been known to terminate sub-contracts to protect the head contract. Reputational and incident exposure: a cyber incident on a 05-138-non-compliant defence system has parliamentary-scrutiny consequences and meaningfully damages the supplier's standing in future MoD competitions. For AI systems specifically, 05-138 overlays the Defence AI Playbook obligations — a customer in scope of both faces compounded contractual exposure if either pack fails.