Mistral Founder Warns: Proprietary AI Models Put Trade Secrets at Risk
Arthur Mensch, founder of French AI company Mistral, warned companies on LinkedIn against depending on proprietary AI models. His argument: closed-model providers increasingly store customer data and gain insight into business processes, and some vendors have historically used such knowledge to compete against their own successful clients. THE DECODER reported the statements on July 5.
His prescription is structural, not just contractual: keep data in open systems, set custom access rules for AI systems, and build independent AI training capability. Mensch contends that self-controlled AI creates advantages competitors cannot replicate, a line that echoes recent remarks by Palantir CEO Alex Karp on enterprise AI sovereignty. The self-interest is obvious: Mistral cannot match the frontier models of OpenAI or Anthropic on raw capability and positions itself squarely on EU sovereignty.
Discount the messenger, keep the mechanism. For German Mittelstand firms whose entire value is process knowledge (the classic hidden-champion profile), the question of what a model provider learns from your usage is not paranoia, it is competitive hygiene. Watch whether procurement teams start writing training-data and insight-access clauses into AI contracts the way they once wrote source-code escrow into ERP deals.