**[Mistral AI Founder Warns: Closed-Source Models May Steal Corporate Data and Take Customers]** According to monitoring by Beating, Mistral AI founder Arthur Mensch issued a stern warning regarding enterprise AI selection. In his view, closed-source large model vendors mandating data retention is a major trap. Once corporate data is integrated, it can be fully accessed and learned by the vendors, granting them significant commercial leverage over the enterprise. He even claimed that closed-source giants have a history of using such sensitive information to directly become the most successful competitors of their own clients across industries. To avoid being trapped in the walled gardens of software giants, he advises enterprises to store their data in open systems. If a vendor refuses to provide full data access, enterprises should leverage AI to swiftly migrate away. Once the data is secured, enterprises must meticulously manage access permissions, combining rigid rule-based systems with flexible large models to prevent employee overreach. The most critical step is to establish a proprietary continuous training flywheel, refining business advantages into exclusive systems that competitors cannot replicate based on interaction feedback, while saving on high deployment costs through model compression. Mensch admitted that this comprehensive reconstruction of IT architecture and development models is highly complex, requiring enterprises to understand both human behavior and gradient descent. To lower implementation barriers, Mistral offers the Studio console and Forge training platform, along with expert teams to guide clients step-by-step through deployment. Once the system is operational, Mistral steps back. He emphasized that while cutting-edge AI can indeed accelerate business growth, if the control switch is not firmly in the hands of the enterprise itself, such growth cannot be converted into the company’s own dividends. [Original Link]