Arthur Mensch

On 16 October 2026 Arthur Mensch published his article Build or surrender, arguing against a position he says is common in Europe: leave frontier models to the Americans and the Chinese. Europe built cloud applications and left the platforms to others, and its companies now pay more than 260 billion euros a year to US cloud and software suppliers. Internet ran on ownerless open protocols; a closed model has an owner who sets prices, restricts uses, and withdraws versions. In June, he writes, a US lab suspended access to its most advanced models, then restored it, without Europe being consulted.

Five months earlier, under oath before the commission d'enquête vulnérabilités numériques, the cofounder and chief executive of Mistral AI compressed the argument into one line: the cloud is artificial intelligence. He set a two-year window before hyperscalers monopolize European energy capacity, and said regulation favors large players. On the engineering side he reported that Mistral's engineers no longer write a line of code and now act as managers of agents.

The unresolved part is where the guardrail sits. Mensch denies a model publisher the democratic legitimacy to arbitrate final use, declining any veto over what the French armed forces do with Mistral's models, yet he also holds that a state must be able to audit, adapt, and guarantee future access to the models it uses in cyberdefense. Shieldstral, published less than three months after that hearing, resolves this only halfway: the classifier carries no taxonomy of licit and illicit, and the deployer writes the question it answers.