The Shared Security Responsibility Model (SSRM) Ownership Map

AI controls fail quietly in the gaps between organisations. Your app calls an orchestrator, the orchestrator calls a model, the model runs on someone else’s cloud — and a control everyone assumes “the platform” handles is owned by no one. The Shared Security Responsibility Model (SSRM) ownership map makes that chain explicit: for every in-scope control, exactly one named owner.

The five seats at the table

• You — the AI customer: scope, data, acceptable use, the final risk decision.
• Application provider (AP) — the product surface where prompts and outputs live.
• Orchestrated service provider (OSP) — the pipeline gluing models, tools and data.
• Model provider (MP) — training, weights, model behaviour and its documentation.
• Cloud service provider (CSP) — the infrastructure everything inherits.

These are the same four provider roles the Cloud Security Alliance (CSA) writes its implementation and auditing guidance for — which is why every control objective in the library drill carries per-role auditing steps. The map and the audit speak the same language.

What the map surfaces

• Ownership gaps — a control no party has accepted in writing.
• Chain mismatches — you assume the OSP filters inputs; the OSP’s terms say you do.
• Justified NAs — controls legitimately owned upstream, which become contract clauses and evidence requests rather than engineering work (feeding the coverage assessment’s NA column).

A shared-responsibility matrix is not paperwork. It is the list of controls you believed someone else was doing.

Using it

The map drives three concrete actions: contract reviews (turn assumed ownership into written ownership), the provider-evaluation questions in the external-assurance artefact, and the unowned-gap entries at the top of the remediation roadmap — because a control nobody owns is the most dangerous kind of NO.

SSRM follows the Cloud Security Alliance (CSA) shared-security-responsibility model as applied to the generative AI service-delivery layers in AI Controls Matrix (AICM) v1.1.0. Ownership shown in the matrix is the typical attribution — your contracts decide the real one, which is precisely why the map exists.