Mistral AI co-founder and CEO Arthur Mensch argues that Europe should invest in its own frontier AI companies. In a Le Monde column dated October 15, 2026, he says proprietary AI models create a different kind of dependence from the open protocols that underpin the internet. Mistral’s own framework makes the idea of AI sovereignty more concrete: control over data, models, computing resources and systems in production.
Why Mensch distinguishes AI from the internet
Internet protocols let different services communicate without one company owning the underlying rules. Mensch argues that proprietary AI models work differently: their owners control the models that shape what applications and agents can do. If an application relies on a closed model, the provider retains control over that capability. That distinction is central to his case for European investment in frontier AI.
What AI sovereignty means in Mistral’s framework
Mistral describes sovereignty through four areas of control in its own framework:
- Data: Information stays within organizational boundaries.
- Models: Organizations can control and customize the models they use.
- Compute: Computing resources are private and predictable.
- Production systems: Deployed systems can be controlled and audited.
This framing makes sovereignty more than a question of where an AI provider is based. Mistral’s framework asks who controls each part of the system, from organizational data to the AI running in production. It is the company’s own description, not a universal standard.
Anthropic’s Fable and Mythos access episode
ITPro reported in June 2026 that Anthropic suspended access to Fable and Mythos for all users after US export controls barred foreign nationals from using the systems. In its October 15 coverage of Mensch’s column, ITPro reported that access had later been restored. Mensch used the episode to argue that decisions about access were made outside Europe.
