Note no. 2 · Digital sovereignty and asset data

Maintenance AI: you keep ownership, you lose control

Nizar Younes Mqam ORCID 0009-0008-5549-7620

Note · 17 September 2026 · Cahiers de technopolitique, no. 2

Every predictive maintenance platform contract says the same thing: your data belongs to you. That is true, and it changes nothing. What is at stake is not the ownership of the raw material, but the ownership of what is drawn from it.

What the contracts grant you

Three providers publish their terms in full and in free access; two of them, Samsara and Siemens, state in so many words who owns the data. The others do not, which is itself a piece of information, and I will come back to it.

Samsara, in its Master Terms of Service updated in June 2026, sets down in its article 8.1 an unambiguous sentence: "Customer owns all Customer Data". Siemens, in its Universal Customer Agreement of 13 April 2026, writes in article 6.3 that "Siemens will not acquire any title to or ownership of Customer Content".

So you are the owner. On this point, neither of them cheats.

And what their definitions take back

The shift is not in the ownership clauses; it is in the definitions, several pages away.

Samsara defines Usage Data as anonymised, de-identified or aggregated data relating to the customer's use of the Samsara products, then includes this Usage Data in the definition of the Services, and finally declares in article 10.1 that "Samsara and its licensors exclusively own all right, title and interest in and to the Services". The loop is closed in three cross-references.

Siemens proceeds differently, more directly. Its article 6.2 allows it to "collect and derive information, statistics, and metrics […] or from Customer Content" and to use them to "develop, and improve its products and services", provided only that what it derives from customer content is aggregated so that the content is no longer identifiable. And its definition of Siemens IP covers "algorithms, models" as well as "any improvement, modification, or derivative work".

The analyses, reports and alerts produced from your data remain filed under your data; what passes to the provider is the aggregated statistics and the improvement of its models. So you keep the ore, and the ingot delivered to you. The foundry's know-how stays with the foundry.

There remains the detail that decides everything, and that one only reads when the relationship breaks down. At Samsara, the licence to exploit the customer's data "shall survive the termination of this Agreement", unless legally prohibited or unless the customer asks in writing, on termination, that it be limited to non-personal data, while that same data "may be immediately deleted". At Siemens, article 10.4 makes article 6.2 survive termination, and the customer has thirty days, provided it complies with the agreement and pays any applicable fees, to retrieve its content, after which it may be erased.

In other words: the day you leave, what you brought disappears on your side and remains on theirs.

European law stops at exactly the same place

The European Data Act has applied since 12 September 2025. It is ambitious: its article 3 requires, for connected products placed on the market after 12 September 2026, access by default, easy, secure, free of charge and in a machine-readable format, to the data generated by their use. Its article 4 requires data to be made available "of the same quality as is available to the data holder", continuously and in real time where relevant and technically feasible.

Its recital 15 closes the door. "Information inferred or derived from such data, which is the outcome of additional investments into assigning values or insights from the data, in particular by means of proprietary, complex algorithms" falls outside the scope of the regulation, unless otherwise agreed between the user and the data holder. The Commission confirms it in its official fact sheet: inferred or derived data is out of scope.

The regulation guarantees the raw material and abandons the finished product.

Two further points complete the picture. The definition of product data, in article 2, covers only data "designed to be retrievable" by the manufacturer: the scope of the right is set by the provider's architecture. And the obligation of functional equivalence when switching provider, in article 30, applies only to infrastructure services. Under this article, software maintenance platforms owe only open interfaces, compatibility with common specifications once these are published and, until then, the export of their exportable data, a definition that precisely excludes what is protected by the provider's intellectual property or trade secrets.

I counted on the full text: the regulation never mentions trained models. Not a single occurrence of "artificial intelligence". A single one of "machine learning", about something else.

The lever exists, but it is contractual

It would be dishonest to conclude that the asset owner is defenceless. The regulation gives it two serious footholds, provided it seizes them.

Article 4, paragraph 13, prohibits the data holder from using non-personal data other than "on the basis of a contract with the user", and forbids it from deriving insights about the user's assets or production methods in a way that undermines its commercial position. Article 13, paragraph 5, presumes unfair a unilaterally imposed term that would allow one party to access and use the other party's data in a manner that significantly harms its legitimate interests.

These two articles do not apply on their own. They have to be argued in court, or better, negotiated before signature. What you do not write into the contract, the regulation will not write for you.

What the Deere case settled, and how

In January 2025, the Federal Trade Commission and two American states, joined in February by three more, sued Deere before the federal court in Illinois. The complaint concerns the Service ADVISOR diagnostic tool, reserved for the dealer network. A settlement was announced in July 2026, in the form of a ten-year order.

The remarkable point is not the outcome, it is the wording of the remedy. It is written in terms of data capabilities: being able to read and clear fault codes, reprogram a control unit, pair a replacement part, take a machine out of degraded mode. Never in terms of ownership, except to specify that access need not include title to Deere's intellectual property.

The same shift can be seen elsewhere. The principles published by IATA in October 2024 on aircraft operational data call for the data to be accessible in the format generated by the aircraft. ISO 55013:2024, which deals precisely with the management of data assets, addresses interoperability and digital preservation, and explicitly excludes any valuation method from its scope.

Three independent sources, in three unrelated worlds. None of them grounds access in title. All of them speak of operational control and format.

What an asset owner can demand

The lesson transfers directly to the specification. Write your requirements in terms of capabilities and formats, not ownership, because ownership will be conceded to you without effort and will bring you nothing.

Four requirements fit in a few lines and change the nature of the contract. Continuous extraction of your raw data in an open and documented format, for the whole duration of the contract. Access not only to the model's outputs, but to the input variables that produce them. A right of export on termination, with a deadline and a format written in black and white, rather than the thirty days by default of a contract you did not negotiate. And explicit limits on the use of derived data outside the performance of the contract, which article 4, paragraph 13, and recital 25 of the European regulation precisely allow you to set.

ISO 55001:2024 gives you the framework: its 2024 edition recast the former clause on information requirements as clause 7.6, "Data and information", which requires data to be specified, along with their quality and source, and their collection, integration and sharing to be planned. Coupled with clause 8.3 on externally provided processes, products, technologies and services, it offers a normative anchor for what would otherwise look like a requirement of mood. The standard provides the framework; the contract provides the teeth.

What we do not know, and what matters

Three providers publish their terms. I found no public contractual text for GE Vernova APM, Schneider EcoStruxure, ABB Ability, SAP APM, Hitachi, nor for Siemens Senseye on its own. The third, AVEVA, is the most direct: its Software Schedule mentions machine learning only to decline to guarantee its outputs, but its general terms grant it, for the term of the contract, a worldwide, royalty-free, sub-licensable licence to use customer content to develop, improve, modify and optimise its products.

The market is measured as poorly as it is documented. Eurostat surveyed European companies on the use of the internet of things for condition-based maintenance only in 2020, in a question that mixed it with tracking the movement of vehicles or products, and then in 2021, in a question of its own: 6.92% of companies with ten or more employees, 21.43% from two hundred and fifty upwards. Its 2025 survey on artificial intelligence, which nevertheless lists many purposes, contains no "predictive maintenance" category.

As for the often announced 25 to 30% reduction in maintenance costs, it does appear in the reference guide of the United States Department of Energy, but it is attributed there to independent surveys that are never named. The figure of 8 to 12% compared with preventive maintenance alone, given in the same document, is no better sourced: it is attributed to earlier studies that are not named.

A market measured this poorly is a market where the buyer negotiates blind. That is precisely where dependence takes root: not in an unfair clause, but in an imbalance of information that no one organised and that everyone puts up with.

Sources and references

All sources were consulted and verified between September and October 2026. The [n] cross-references in the text point to this list.

1. Regulation (EU) 2023/2854 of 13 December 2023 on harmonised rules on fair access to and use of data (Data Act). EUR-Lex, CELEX 32023R2854. Articles 2, 3, 4, 13, 29, 30, 31 and 50; recitals 15 and 25.
2. European Commission, Data Act explained, digital-strategy.ec.europa.eu.
3. Samsara, Master Terms of Service, June 2026 update, articles 8.1, 10.1, 18.29 and 18.34.
4. Siemens, Universal Customer Agreement version 1.3 of 13 April 2026, articles 3.6, 6.2, 6.3 and 10.4, and definition of Siemens IP.
5. AVEVA, General Terms and Conditions version 6.1 of 16 July 2026 (version 6.0 of 31 March 2026), article 13, and Software Schedule P&O APM M&C version 6.0, June 2026, article 3.5.
6. Federal Trade Commission and five states v. Deere & Company, United States District Court for the Northern District of Illinois, case 3:25-cv-50017, complaint of 15 January 2025 (FTC, Illinois, Minnesota), amended on 7 February 2025, settlement announced on 8 July 2026. Court approval had not been confirmed as of 17 September 2026.
7. IATA, principles on aircraft operational data, 2 October 2024. GE, Pratt & Whitney and Boeing are not signatories.
8. Eurostat, series isoc_eb_iot, indicator E_IOTDMTN1, 2021 survey.
9. Eurostat, series isoc_eb_ai, 2025 survey.
10. United States Department of Energy, Operations and Maintenance Best Practices Guide, version 3.0, PNNL-19634, 2010, section 5.4.
11. ISO 55001:2024, Asset management, Asset management system, Requirements, second edition, July 2024, clauses 7.6 and 8.3.
12. ISO 55013:2024, Asset management, Guidance on the management of data assets.

Download

Read the document (PDF)

5 pages · English · 12 sources · 87 KB
Fully selectable and citable text
The document carries its own DOI, reserved before export
CC BY-NC-ND 4.0 licence, no registration

Bibliographic record

Title

Maintenance AI: you keep ownership, you lose control

Author

Nizar Younes Mqam, ORCID 0009-0008-5549-7620

Serial publication

Cahiers de technopolitique, no. 2, Note format

ISSN

3131-7979

Publisher

Nizar Younes Mqam, Tourcoing, France

Publication date

17 September 2026

Revision

7 October 2026: every quotation, figure and attribution checked against its primary source; dates, figures, attributions and references corrected, cuts in quotations marked [...]. The argument is unchanged.

Language

English

DOI

10.5281/zenodo.23207206
All versions : 10.5281/zenodo.23199125

Permanent repository

Zenodo, hosted by CERN

Licence

CC BY-NC-ND 4.0

Other languages

Français (original) · العربية

To cite this study

Mqam, N. Y. (2026). Maintenance AI: you keep ownership, you lose control. Cahiers de technopolitique, n° 2. https://doi.org/10.5281/zenodo.23207206

Read also

No. 7 · Study · October 2026

ISO 55001 and the AI agent: who signs before the auditor?

The next step: once the maintenance AI becomes an agent that acts, the question is no longer only who owns the data, but who answers for the decision.

11 pages · 30 sources · PDF 115 KB
DOI 10.5281/zenodo.23207212

Also inFrançaisالعربية

Read the record and download

No. 3 · Study · September 2026

Asset data: exit becomes free, the model stays captive

Asset data: exit becomes free, the model stays captive. The Data Act makes switching provider free from 2027; what you can take with you is another matter.

14 pages · 28 sources · PDF 125 KB
DOI 10.5281/zenodo.23207208

Also inFrançais

Read the record and download

No. 1 · Study · September 2026

When the cloud catches fire

When the cloud catches fire: the platforms that host maintenance AI are physical buildings, and in March 2026 drones reached them.

22 pages · 43 references · PDF 177 KB
DOI 10.5281/zenodo.23270771

Also inFrançais

Read the record and download