Where the Cloud Ends and Your Duty Begins: A 2026 UK Shared Responsibility Guide
Photo: Colin Smith, CC BY-SA 2.0, via Wikimedia Commons
Cloud providers secure a great deal, but they do not run your IT department. That distinction sounds obvious until an incident occurs. A UK organisation discovers that a storage bucket was publicly accessible, a SaaS tenant lacked multi-factor authentication, or a backup had never been tested. The provider's infrastructure was resilient. The customer's configuration was not. The shared responsibility model explains why both statements can be true at the same time.
In 2026, the model is more important than ever because UK organisations consume cloud through a mixture of infrastructure, platform and software services. Each service type moves the line. A decision-maker who assumes the provider covers everything will eventually be disappointed. A decision-maker who assumes the provider covers nothing will overspend on duplicated controls. The practical task is to know where the line sits for each workload and to evidence who does what.
The line is not a single line
Under infrastructure as a service, the provider secures the physical data centre, hardware, network fabric and hypervisor. The customer remains responsible for the guest operating system, patching inside that system, application configuration, identity controls, network rules, data classification and backup. The provider may offer backup tooling, but the customer must design, schedule and test it.
Under platform as a service, the provider takes on more of the runtime and managed service layer. The customer still owns the application, its configuration, access controls and data. A managed database service may handle patching of the database engine, but the customer must still manage schema changes, permissions, backups where required, and the classification of the data held within it.
Under software as a service, the provider operates the application and underlying platform. The customer owns user identities, access permissions, sharing settings, data classification, retention decisions and often backup of SaaS data. Provider resilience is not the same as customer backup. Replication across data centres protects service availability; it does not necessarily restore a file deleted by a user or a record overwritten by a faulty integration.
Where UK organisations commonly misread the gap
Patching. In IaaS, the customer patches the guest operating system and middleware. In PaaS, the provider patches the managed runtime, but the customer patches application dependencies. In SaaS, the provider patches the application, but the customer must manage integrations and client software. The confusion arises when teams assume that because a service is managed, every layer is managed.
Identity. Providers secure the identity platform itself. Customers secure the accounts, roles, conditional access policies, privileged access and joiner-mover-leaver process. A provider can offer strong identity tooling, but it cannot decide which of your staff should hold administrator rights.
Backup. Providers design for durability and availability. Customers design for recovery. Those are different objectives. A UK SME may need point-in-time restore for Microsoft 365 data, a tested restore process for a line-of-business application, or an isolated recovery environment for ransomware. The provider's service status page will not demonstrate that the customer can recover.
Data classification. The provider does not know which records are personal data, commercially sensitive, legally privileged or subject to UK GDPR retention requirements. Classification, labelling, data loss prevention and retention rules remain customer responsibilities. So does the decision about which data may leave the UK or be processed by a subprocessor.
Incident response. Providers detect and respond within their own environment. Customers must detect and respond within their tenant, application and identity estate. Logging, alerting, escalation paths and forensic readiness need to be designed jointly, not assumed.
A practical decision matrix
| Responsibility area | IaaS | PaaS | SaaS | Evidence to request |
|---|---|---|---|---|
| Physical data centre and hardware | Provider | Provider | Provider | Audit reports and certifications |
| Hypervisor and managed runtime | Provider | Provider | Provider | Service description and patching statement |
| Guest operating system patching | Customer | Not applicable or shared | Not applicable | Patch reports and vulnerability scans |
| Application patching and dependencies | Customer | Customer | Provider for core app; customer for integrations | Vulnerability management reports |
| Identity configuration and privileged access | Customer | Customer | Customer | Access reviews and conditional access evidence |
| Data classification and retention | Customer | Customer | Customer | Data map, labels and retention schedule |
| Backup design and restore testing | Customer | Customer, unless explicitly managed | Customer for SaaS data unless contracted | Restore test records and RPO/RTO agreement |
| Monitoring and incident response | Shared | Shared | Shared | Runbooks, alert coverage and escalation contacts |
| Network configuration and firewall rules | Customer | Shared | Shared | Configuration baseline and change records |
| UK data residency and subprocessors | Provider options; customer selects | Provider options; customer selects | Provider options; customer selects | Data residency statement and subprocessor list |
Questions to put to your MSP or cloud account team
- Which controls are explicitly assigned to the provider in the contract, and which remain with us?
- Can you provide a responsibility matrix for each service we use, not just a general model?
- What evidence do you hold for patching, vulnerability management and configuration baselines?
- Which identity controls are included, and which require our configuration or an additional licence?
- What are the backup, restore and recovery objectives for each service, and when were restores last tested?
- For SaaS, who is responsible for point-in-time restore, and what is excluded?
- How do you support data classification, labelling and UK GDPR retention requirements?
- What are the incident notification timelines, escalation routes and forensic responsibilities?
- Which subprocessors handle our data, and where is data stored or replicated?
- What are the exit terms, data export formats and deletion timescales at contract end?
Turning the model into an operating rhythm
The shared responsibility model should not live in a slide deck. It should be attached to the contract, reflected in service descriptions and reviewed alongside incidents and audits. UK decision-makers should ask for a named owner on both sides for each major service, a quarterly review of control evidence and an annual test of the incident and restore processes.
The provider supplies the platform. The organisation remains accountable for how that platform is configured, who can access it, what data it holds and whether recovery has been proven. In 2026, that is the practical meaning of shared responsibility: not a disclaimer, but an operating discipline.