Mistral criticizes US control over AI models

What happened?

On May 12, 2026, the United States government issued an executive order directed at Anthropic, one of the leading artificial intelligence companies, requiring it to restrict access to its most capable models — such as Claude 4 Opus — to non-US citizens. The measure, based on national security and control of dual-use technologies, clashed with technical reality: Anthropic's systems cannot discriminate by nationality without compromising privacy or functionality. Faced with the impossibility of implementing the order selectively, the company temporarily withdrew access to these models worldwide, affecting users in all countries. This incident was denounced by Arthur Mensch, CEO of Mistral AI, during a G7 meeting in Evian, where he shared a table with Sam Altman (OpenAI), Demis Hassabis (Google DeepMind), and Dario Amodei (Anthropic). Mensch described the situation as a 'sovereignty moment' and highlighted Europe's dependence on US-controlled infrastructure, noting that Washington's unilateral decision had left European companies and researchers without access to key AI tools.

Why is this important?

This incident highlights the fragility of technological autonomy in regions like Europe in the face of unilateral decisions from Washington. This is not the first time the US has used export controls to limit access to critical technologies: from restrictions on NVIDIA chips to China in 2022, to sanctions on Huawei, the pattern is clear. However, the novelty is that AI models themselves are now the target, not just hardware. Mistral AI has been warning for two years that the US could 'shut down' its international competitors through export or access controls, and this case confirms its fears. The order against Anthropic, though aimed at a specific target, affects users worldwide and sets a dangerous precedent: any government with power over cloud infrastructure or AI providers could impose similar restrictions. Moreover, it underscores the urgency of developing sovereign AI models and own ecosystems, such as the one Mistral promotes with its Le Chat assistant. According to European Commission data, 80% of AI models used in Europe come from US providers, exposing the region to risks of strategic dependence.

Consequences for the sector

• Market fragmentation: AI models could be split into national or regional versions, losing the advantage of global scale. This would increase development costs and limit cross-border collaboration in research.
• Race for sovereignty: Countries like France, Germany, and China will accelerate investments in their own AI to reduce dependencies. For example, France has already announced a €5 billion fund for sovereign AI, and Germany plans to create a European consortium for foundational models.
• Impact on startups: Non-US startups relying on APIs from OpenAI or Anthropic will face regulatory uncertainty. A Dealroom study indicates that 60% of European AI startups use US APIs, making them vulnerable to political changes.
• Ethical debate: The measure reopens discussion on government control of AI and universal access rights. Organizations like Access Now have criticized the order for violating principles of equal access to knowledge.

What should readers know?

The US order is not an isolated event. It is part of a broader strategy of technological control that includes restrictions on chip exports (such as limitations on NVIDIA A100 and H100 to China), semiconductor design software, and now AI models. For European companies and developers, it is crucial to diversify suppliers and consider alternatives such as open-source models (e.g., those from Meta or Mistral) or European platforms. Mistral, with its focus on sovereignty and open source, positions itself as a viable option, but it still needs to scale to compete with US giants: its most powerful model, Mistral Large, has 70 billion parameters, compared to over 1 trillion for Claude 4. This 'sovereignty moment' could catalyze a new wave of investment in regional AI. However, experts warn that technological autonomy is not achieved solely with own models; cloud infrastructure, local data, and talent are also needed. The open question is whether Europe can turn this crisis into an opportunity to build an independent and competitive AI ecosystem.

'The order against Anthropic is a wake-up call for Europe. We cannot allow our innovation capacity to depend on decisions made in Washington.' – Arthur Mensch, CEO of Mistral AI.