The digital world produces announcements every week about new AI models, strategic partnerships and spectacular funding rounds. Behind this constant stream, one question is becoming increasingly important: among the players issuing one press release after another, which can prove that they truly control their models, data and energy footprint—and that their customers can switch providers?
Digital sovereignty for businesses: what the first indices measure
Digital sovereignty is about more than hosting data within national borders. It also depends on an organisation’s practical ability to control its AI systems and migrate to another provider if necessary. IBM has published an initial Digital Sovereignty Index focused on France, which seeks to put this operational reality into measurable terms.
This kind of initiative remains a corporate study, with all the potential biases that entails. Its value lies in its approach: rather than comparing political statements about sovereignty, the index assesses actual practices (data portability, model governance and contractual reversibility). This is an area where the latest news on E-novateur can help track concrete developments across the sector.
| Sovereignty criterion | What is claimed | What can be measured |
|---|---|---|
| Data hosting | Data stored within national borders | Verifiable contractual clause, SecNumCloud certification or equivalent |
| AI model governance | Algorithmic transparency | Technical documentation, access to training logs |
| Reversibility (switching providers) | Guaranteed portability | Open formats, tested migration timelines |
| Environmental footprint | Claimed carbon neutrality | Published, auditable consumption data |
The gap between the “claimed” and “measurable” columns remains the main blind spot. Most cloud providers communicate where their servers are located, but far less about genuine portability or model traceability.

AI Act and regulatory sandboxes: Europe’s timeline tightens
The European regulatory framework for artificial intelligence is advancing in stages. An implementing regulation adopted by the European Commission on 10 November 2026 and published in the Official Journal the following day sets common rules for establishing, operating and supervising AI regulatory sandboxes.
These testing environments allow companies to trial AI systems under the supervision of a competent authority before bringing them to market. The regulation harmonises access conditions across Member States.
The timeline for some AI Act obligations has also been revised. Recent additions include a ban on certain uses that generate non-consensual intimate content or child sexual abuse material. These prohibitions strengthen the regulation’s protective dimension, extending beyond economic oversight alone.
What this means for developers and businesses
- Companies developing high-risk AI models must prepare to enter a regulatory sandbox, with criteria now harmonised at European level.
- The technical documentation required to enter a sandbox (training data, bias assessments and safety measures) provides an initial test of the system’s actual maturity.
- The prohibitions added to the AI Act timeline create a blacklist of uses that every digital-sector player must take into account from the design stage.
AI and data centre environmental footprints: the problem of missing figures
On 9 November 2026, Arcep brought together industry stakeholders in Brussels to discuss the digital sector’s environmental footprint. The authority called for coordinated action at European level. At the same time, its press review reports that precise data on energy consumption at French data centres remains difficult to obtain.
This creates a paradox. On the one hand, cloud providers and data centre operators are making ever more carbon-neutrality commitments. On the other, regulators are struggling to verify these claims because granular data is not being published.
Energy transparency: a factor in business decisions
For a company choosing an AI infrastructure provider, environmental transparency is becoming an operational criterion. A provider able to supply consumption figures by service, model or query offers a level of visibility that most still do not provide.
Arcep’s call highlights a structural issue: without standardised metrics at European level, each player defines its own indicators, making comparisons impossible. The implementing regulation on regulatory sandboxes could inspire a similar approach to environmental impact by establishing common reporting frameworks.

Regulating consumer AI assistants: user vulnerabilities under scrutiny
A French commission has recommended tighter regulation of consumer AI assistants, from ChatGPT to Gemini and Grok. Its recommendations focus on protecting vulnerable users, particularly minors and people experiencing psychological distress.
At this stage, these recommendations are not binding. They outline areas of responsibility that AI assistant providers will probably need to address: appropriate notices, limits on conversations about certain sensitive topics, and mechanisms for directing users to human support resources.
The issue goes beyond technical considerations. An AI assistant responding to a distressed teenager or an isolated person raises questions about its provider’s responsibility that current terms of service do not address. The gap between these tools’ conversational capabilities and the lack of appropriate safeguards is fuelling a debate that is set to take shape over the coming months.
Digital and innovation news is often viewed through the lens of performance (faster models, more powerful chips). The regulatory and environmental signals of November 2026 shift the focus towards demonstrable control: control over data, energy impact and interactions with vulnerable groups. The credibility of future announcements will be decided on this front.



