Mistral's open-source pitch: closed AI models let labs spy on your business
mistral ceo arthur mensch claims proprietary models let labs spy on business processes and steal alpha, pushing open-source self-hosting as the only way to ensure data sovereignty.
arthur mensch, ceo of mistral ai, issued a warning against relying on closed-source models, asserting that proprietary providers accumulate customer data to reverse-engineer business processes and target successful clients. he urges companies to deploy open systems and self-hosted models to maintain sovereignty over their institutional knowledge, though critics note his stance aligns with mistral's eu-focused business model against us giants. his argument is bolstered by a recent experiment from bridgewater and thinking machines lab, where a fine-tuned qwen model achieved 84.7% accuracy on financial documents compared to a frontier model's 78.2% at a fraction of the cost. however, closed-source labs could eventually acquire or generate similar proprietary data to close the gap.