One word, several requirements: what sovereign cloud covers
Sovereign cloud became a sales argument before it became a precise notion. Very different offerings claim it: an independent European host, the European region of a global provider, an offering operated by a company under French law using licensed technologies. In the absence of a single legal definition, the word guarantees nothing by itself.
Projects that get bogged down start from this ambiguity. Sometimes the organisation demands “sovereign” for everything, then backs away from the cost and the narrowness of certain catalogues. Sometimes it puts the question off until later, until the day an audit, a regulator or a customer puts it to the organisation in precise terms.
The expression “trusted cloud” (“cloud de confiance”) is used by the French public authorities: the “Cloud de confiance” label is one of the three pillars of the national cloud strategy announced on 17 May 2021. In both cases, the right question is not “is this offering sovereign?”, but “what guarantees does it provide, and which ones do I need for this use?”.
Location, applicable law, control: three distinct questions
Where is the data? Location concerns storage, but also processing, backups, logs and support data. For personal data, transfers outside the European Union are governed by Chapter V of the GDPR, Regulation (EU) 2016/679.
What law is the provider subject to? A provider that falls under the law of a third country, directly or through its parent company, may receive an access request from that country’s authorities, including for data stored in Europe. This exposure depends on the structure of the company and its shareholding, not on the address of the data centre.
Who operates, and who holds the keys? Administration staff, support tools and encryption key management determine who can technically access the data. Keeping your keys outside the provider reduces exposure. It does not eliminate it, since the data is decrypted at the time of processing.
These three questions call for three separate answers. An offering may satisfy the first and not the second.
SecNumCloud: what the qualification says, and what it does not
SecNumCloud is ANSSI’s qualification framework for cloud service providers. Four reading precautions are essential.
- The qualification applies to a service, not a company. A provider may offer a qualified service alongside many services that are not. The exact scope is specified by ANSSI in the list it publishes.
- The source of truth is the list published by ANSSI. A sales announcement, a process “under way” or a certification of another kind does not amount to qualification. The status of an offering is checked on this list, on the date of the decision.
- The framework evolves. Its version 3.2, dated 8 March 2022, incorporated criteria for protection against non-European law (chapter 19.6 of the framework).
- Qualification of the service does not mean compliance of your system. Configuration, access management and the security of your applications remain your responsibility.
Other schemes coexist without being the same. Health Data Hosting (HDS) certification, provided for in Article L. 1111-8 of the French Public Health Code, meets an obligation specific to health data. It does not entail SecNumCloud qualification, and the reverse is not true either.
Hybrid architectures: distributing without siloing
A large information system almost always combines several environments. The question is according to what principle.
| Model | Principle | Suitable when | Price to pay |
|---|---|---|---|
| Distribution by sensitivity | Sensitive workloads go to a qualified or internal environment, the others to a public cloud | Data classification is reliable and flows between classes are controlled | Two foundations to operate, gateways to secure |
| Extension of the data centre | The internal environment remains the foundation, the cloud absorbs variable or new needs | The internal estate is recent and well operated | Demanding network interconnection, dual skills |
| One cloud per application | Each application lives in a single environment, chosen according to its needs | Applications are loosely coupled | Scattered practices if no common framework exists |
| Active portability | The same application can be deployed on several environments | The continuity or reversibility of this application justifies it | Giving up services specific to each provider, testing regularly |
Whatever the model, three functions must remain common: identity management, monitoring and the way of deploying. Without them, hybrid becomes a juxtaposition of silos, each with its own tools and blind spots.
Portability, finally, has a cost. Sticking to open standards (containers, orchestration with Kubernetes, open source databases) makes it easier to change environment, but rules out managed services that are sometimes very useful. We recommend setting the level of portability application by application, instead of making it a general rule. The contractual and operational aspects of exit are dealt with in our page on cloud migration.
Our approach: classify before choosing
We proceed in four steps, in this order.
Classifying data and processing. Without classification, any discussion about the provider is premature. This work builds on your data governance where it exists, and moves it forward where it does not.
Setting a requirements grid. For each class, the grid specifies location, applicable law, the expected qualification level, encryption and exit conditions. It is validated by security, legal and the business.
Matching workloads against the grid. Each application is given a hosting target. Exceptions are named, justified and given a review date.
Designing the architecture and preparing the contract. Interconnection, identities and monitoring are designed for the whole. The technical requirements are translated into clauses that your lawyers negotiate.
For entities subject to resilience obligations, this approach ties in with the management of third-party risks, described in our page on cyber resilience.
The misconceptions to avoid
- Confusing address and jurisdiction. A data centre located in France says nothing about the law to which its operator is subject.
- Placing everything at the most demanding level. The cost and constraints of this level then weigh on data that does not need it, and teams look for workarounds.
- Forgetting the periphery. Administration tools delivered as a service, telemetry, support and backups sometimes leave the protected perimeter without anyone having decided so.
- Reading a qualification as a company label. Only the service designated by ANSSI, within the scope it indicates, is covered.
- Promising portability that has never been tried. An application said to be portable that has never been deployed elsewhere is not yet portable.
How to know whether the trajectory is the right one
The indicators we track with you are simple to observe:
- the share of data and processing actually classified;
- the share of applications hosted in accordance with the grid, and the number of open exceptions;
- the age of exceptions, a sign that they are being dealt with or forgotten;
- the number of non-substitutable services identified per critical application;
- the completion of portability or data return tests, and their results.
A sound trajectory can be recognised by the fact that every hosting decision can be justified in writing, before an auditor as before management.