A company rents its servers from a cloud provider, runs packaged software whose code it does not hold, entrusts developments to service providers and keeps its data in its vendors’ formats. None of these dependencies is a fault: that is how an information system is built. A company’s digital sovereignty becomes an issue when one of them can no longer be undone within a timeframe and at a cost that management knows.
The expression is one of common usage; we know of no single legal definition of it. Parliamentary work defines it for a State: a report by a French Senate committee of inquiry (No. 7, 2019-2020, “Le devoir de souveraineté numérique”) sees it as “the State’s ability to act in cyberspace”. For a company, we propose to understand it as the ability to decide on its digital dependencies. An executive committee can examine it by asking what and whom the business depends on, what would remain if a dependency ceased, what needs protecting first, and at what price. None of these questions is limited to the cloud: software, data, skills and suppliers fall within it just as much.
What does the business really depend on?
The expected answer is an inventory, drawn up essential activity by essential activity. It covers several families of dependencies:
- infrastructure: hosting, network, cloud services, including managed services specific to a provider;
- software: licensed packaged software, software as a service, open source components maintained by third parties;
- data: where it resides, the format in which it can be recovered, the keys that encrypt it;
- skills: the employees who know how to operate and develop each building block, and the service providers who hold this knowledge in their place.
This inventory extends the map described on our IT architecture page. It adds one column to it: the third party on which each component depends. Two dependencies may escape it, because they appear on no hosting invoice: the software-as-a-service tool subscribed to by a business department, and the person, internal or external, who alone knows an old application.
Whom do we depend on, and what law are these third parties subject to?
Behind each line of the inventory stands a chain: the supplier, its parent company, its subcontractors, the vendor of a component it integrates. The committee first examines it from a commercial standpoint: can the third party change its prices or licence terms, or withdraw a product, and with what notice?
Then comes the legal standpoint: what law is the third party subject to? Our entry on the Cloud Act recalls that this United States law reasons in terms of the supplier’s possession, custody or control of the data, whether or not that data is located on American soil. Applying it to a given supplier is a lawyer’s job, which this article does not do.
Then there is operation: who administers the service, and who holds the encryption keys?
The answers can be read in the contracts and in the suppliers’ documents. For a cloud service, a public benchmark exists: SecNumCloud, ANSSI’s qualification framework for cloud service providers. Qualification is granted service by service, and the list published by the agency makes it possible to check it. An offer presented as sovereign is examined on the same documents. Our sovereign and hybrid cloud page details the checks to be carried out on location, applicable law and control of operations.
Dependency on people is examined in the same way. How many employees know how to restart, modify or take over a given application? What does the service provider’s contract provide for regarding documentation and skills transfer?
What would remain if one of these dependencies ceased?
For each critical dependency, the committee asks for a written scenario. Several events can lead to it: failure or acquisition of the supplier, discontinuation of a product, a price rise that makes continuing untenable, termination, departure of the people who held the knowledge, a decision by an authority. The scenario says what stops, how long the business holds out in degraded mode, what replaces the component and within what timeframe.
A workable exit requires data that can be extracted and read without the original tool, a record of the components that only this supplier offers, and a test. As long as data recovery has not been tested, the exit time is only an estimate.
Union law deals with part of the subject. Chapter VI of the Data Act, Regulation (EU) 2023/2854, governs switching between providers of data processing services, a category that covers cloud computing services. For financial entities, Regulation (EU) 2022/2554, known as DORA, requires in its Article 28 exit strategies for services supporting critical or important functions. Our article on multicloud strategy examines these texts, as well as the separate question of the number of suppliers. Keeping the business running during the transition falls under cyber resilience.
What should be protected as a priority?
Bringing all dependencies up to the highest level of control would make data and tools that do not need it pay for that control. The priority is set by classifying data and processing: trade secrets and know-how, personal data, data whose loss would halt the business, data that a contract or a legal text requires to be hosted in a certain way. This classification is for the business functions and relies on data governance, which designates an owner for each domain.
The State reasons in this way for itself. The “cloud au centre” doctrine applies, according to its text, to State bodies and to organisations under its supervision. In its 2023 version, it requires a commercial cloud offer to hold the SecNumCloud qualification, or a European qualification of at least equivalent level, only for data “of particular sensitivity” whose breach is likely to cause the harms that the text lists: otherwise, using such an offer “is not required”. A company that does not fall within this scope can take away its method: first define the category that justifies the constraint, then apply it to that category alone.
The classification also applies beyond data. It designates the applications whose code the company wants to keep control of and the knowledge it is keen to retain in-house.
What price are we prepared to pay for the company’s digital sovereignty?
Each degree of control has a cost, which is named before it is quantified: a narrower catalogue of services, a second platform to operate, skills to recruit or maintain, managed services given up in order to remain portable, exit tests to fund. On the other side, the cost of dependency is that of the cessation scenario: downtime, replacement time and budget.
The committee therefore decides dependency by dependency. It can reduce it: replace the component, bring a skill back in-house, add a second supplier. It can frame it with reversibility clauses, documentation, skills transfer, a periodic exit test. Finally, it can accept it in writing, with a named owner and a review date: acceptance is a legitimate outcome as long as it is explicit.
The minutes of the session record, for each dependency judged critical, the outcome adopted, the person who owns it and the date of the next review. The company’s digital sovereignty can then be read in this register, in the proportion of lines whose exit has been tested.