The recent US decision to suspend foreign nationals' access to Anthropic's Fable 5 and Mythos 5 models under export control rules marks a turning point for the legal artificial intelligence market.

It is a reminder of an obvious truth that is still too often underestimated: choosing an AI platform is not simply choosing software. It means choosing critical infrastructure that will gradually take a central place in day-to-day work, value creation and an organisation's technological trajectory. Choosing dependencies wisely is more essential than ever.

Law is not an activity like any other. It organises economic life, secures human relationships, and frames and legitimises public and private decisions. Legal professionals hold a substantial share of organisations' sensitive information. The digital infrastructure used to produce, analyse and process the law therefore cannot be regarded as mere software, because it safeguards our most fundamental freedoms and rights.

For law firms, in-house legal teams, notaries and all legal professionals, the issue is no longer hypothetical. Internal knowledge is structured in tools, prompts circulate between teams, workflows become professional habits and confidential documents gradually underpin day-to-day productivity. When access to a model can be suspended by an extraterritorial decision, more than a single feature disappears. The entire legal production chain can be weakened.

At Jimini AI, we chose to build sovereign infrastructure that is model-provider agnostic and designed and co-developed with legal professionals. Legal AI cannot rely exclusively on a single provider, however capable it may be. It must support multiple models and be reversible, auditable and controlled, guaranteeing every organisation continuity, data protection and the freedom to evolve its technology stack.

This requirement is all the more important because civil law cannot be treated as a mere local adaptation of a product designed primarily for common law. The reasoning, practices, workflows, standards of documentary rigour and professional conduct constraints of European professionals require a dedicated approach. Jimini is not building a general-purpose AI layer applied to law. Jimini is building European legal infrastructure with national and European legal systems as its starting point.

The debate on legal AI can no longer be confined to model performance. It must now include control over infrastructure, continuity of use and technological sovereignty. Legal professionals have always been the guardians of freedoms and legal certainty. They must now ensure that the tools assisting them meet the same standards.

This is precisely what Jimini AI is building alongside law firms, notaries and in-house legal teams: sovereign, controlled infrastructure designed for the requirements of European law.

Raphaël Arroche, CEO and co-founder, and Stéphane Béreux, CTO and co-founder