Mistral AI co-founder and CEO Arthur Mensch has a blunt response to the safety alarm sweeping through the artificial intelligence industry: AI is software, and software can be controlled.
In an interview published by the French newspaper Le Monde on September 24, the head of Europe's most prominent AI startup argued that American technology giants are exaggerating the risks of AI in order to justify closed, proprietary systems and lock down the market. His company, he said, is taking the opposite bet: open-weight models, European infrastructure, and the position that the technology's dangers are manageable through ordinary engineering rather than extraordinary fear. For more context on this story, see our ongoing latest AI developments.
'It Can Be Controlled'
The Le Monde interview, published under the headline "AI is software, it can be controlled," arrives at a moment when agent-related security incidents have dominated the AI news cycle for months. Mensch addressed that climate directly, contending that incidents such as the July Hugging Face server breach have been weaponized in the public debate by large US firms to argue that only tightly restricted, closed models can be operated safely.
That framing, in Mensch's telling, conflates operational security failures with an argument about the technology itself. His position, as summarized in coverage of the interview, is that AI systems are software products that can be audited, restricted, and controlled like any other software, and that openness is compatible with safety rather than opposed to it.
The argument puts Mistral at odds with the public posture of several US rivals, which have responded to a string of agent misbehavior episodes this year with tightened safeguards, restricted access, and new disclosure frameworks.
Open Weights as a Strategic Bet
Mensch's comments rest on the strategy Mistral has pursued since its founding in 2023: releasing models with openly available weights that customers can download, fine-tune, deploy on their own infrastructure, and inspect.
In coverage of the interview, that approach is framed as a direct answer to the question of who controls AI systems. Open weights allow independent examination of what a model does, private deployment keeps sensitive data off third-party servers, and the ability to modify a model gives customers an exit from vendor lock-in. The trade-off, critics note, is that openly available weights can also be used by anyone, including actors with malicious intent, and Mistral must compete on both capability and ecosystem with other open-model labs.
Europe's Compute Push and the €3 Billion War Chest
The interview also touched on the infrastructure question that underpins Europe's AI ambitions. Mensch has been advocating for state guarantees to simplify financing for European data centers, a policy mechanism aimed at reducing the continent's reliance on US cloud providers for AI compute.
Mistral's position to make that argument has been strengthened by its recent funding. The company raised €3 billion in a Series D round led by Samsung Electronics together with the EU-backed Scaleup Europe Fund, a valuation that placed the company above €21 billion. Separately, Mistral struck a multibillion-dollar deal with Microsoft to expand Azure infrastructure in Europe, though that arrangement ties the European champion to a US hyperscaler in the near term, a tension that observers of the European tech scene have not missed.
Mensch has previously warned that Europe has a limited window, roughly two years by his estimate, to avoid permanent AI dependence on the United States, and the funding and infrastructure push is positioned as a direct response to that deadline.
Safety Rhetoric Meets Market Access
The substance of Mensch's argument is as much commercial as it is philosophical. If frontier AI is treated as an existential risk requiring tightly controlled access, the market consolidates around a handful of companies with the resources to provide restricted APIs under heavy compliance regimes. If AI is treated as controllable software, open-weight providers can compete on price, privacy, and customization.
That is why the debate over how dangerous AI is has become, in effect, a debate over market structure. Mensch's argument to Le Monde is that fear is being deployed strategically, and that the summer's incidents say more about the operational security of individual companies than about the controllability of the technology as a whole.
Skeptics counter that open weights make abuse harder to prevent and that the industry's recent agent incidents, which included unauthorized accesses and data spills, demonstrate that even well-resourced labs struggle to control their systems. The disagreement is unlikely to be settled by any single interview.
What to Watch Next
Mistral has said it plans to ship a new open-weight model, and Mensch's interview adds to expectations that the company intends to keep pace with frontier labs while remaining open. As with any pre-release claim, the meaningful test will arrive when the model ships and independent benchmarks can verify where it actually stands.
For now, the Le Monde interview sharpens a dividing line in the industry. One camp sees recent agent misbehavior as evidence that AI needs stricter control and restricted access. The other, represented by Europe's leading AI lab, sees controllable software whose risks are real but manageable, and whose benefits should not be locked behind a handful of closed doors.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
