Arthur Mensch, founder and CEO of Mistral, is urging companies to think carefully before placing core business work inside proprietary AI systems. His argument is not only about model performance. It is about who sees the data, who sets the rules, and who ultimately benefits when AI becomes part of a company's operating process.
In a LinkedIn post, Mensch warns that closed model providers are collecting increasing amounts of customer data. That, he argues, can give AI labs a direct view into how their customers operate.
The warning from Mistral's CEO
Mensch's central concern is that proprietary AI models do not simply answer prompts. When companies depend on them for important workflows, the model provider may learn about business processes, internal methods, and operational priorities.
According to Mensch, companies that sell closed models are storing more and more data. He claims this gives them a window into their customers' business processes. He also says some AI labs "have a track record of going after their most successful customers thanks to this information," according to the source article.
His proposed answer is for companies to keep more control over their AI stack. Mensch advises businesses to store data in open systems, define their own access rules for AI, and build their own training models, even if "these efforts might seem daunting."
"Frontier AI can accelerate the growth of your business, but if it's not in your hands, it's not going to be your growth,"
The message is direct: AI can become a growth engine, but only if the company using it controls enough of the data, access, and model development process to keep the value from flowing elsewhere.
Why control over AI models matters
The debate is partly about security, but it is also about business leverage. If an outside AI provider handles sensitive workflows, the provider may learn which processes matter, where the customer has an edge, and what kind of knowledge is most valuable.
That is why Mensch frames open-source AI and open systems as strategic infrastructure rather than just a technical preference. In this view, model weights, training choices, and data governance are not back-office details. They shape whether a company can keep its accumulated knowledge inside its own walls.
Palantir CEO Alex Karp has made a similar argument. He has also urged companies to build their own AI models instead of relying on proprietary outside solutions. Palantir published a manifesto for secure AI in business that makes the point in especially blunt terms.
"Controlling your weights is controlling your fate. Weights are the distilled form of hard-won, accumulated institutional knowledge. If you let others control your weights, you are allowing them to migrate the alpha of your business to theirs."
For companies adopting AI, the practical question is no longer only which model performs best on a task today. It is also who controls the system once AI becomes deeply embedded in daily work.
The business context behind the argument
Mensch's position has weight, but it also comes from a company with its own incentives. Mistral is described in the source article as the only EU company with relevant AI models. The same article says it cannot really compete with top-tier models like GPT-5.6 Sol or Fable 5 on raw performance.
That context matters because Mistral's business model leans heavily on EU sovereignty, where the company stands to gain the most. The source article also notes that about 30 percent of its shares are held by US investors.
So Mensch's warning should be read in two ways at once. It is a serious argument about data, control, and competitive risk. It is also an argument that supports the market position of the company he runs.
The performance picture is also mixed. The source article says large general-purpose AI models have repeatedly beaten specialized models on specialized benchmarks, as long as the relevant domain knowledge was included in the training data. That weakens any simple claim that company-built or specialized models will always be better.
Where open models may have an edge
There is still evidence that company-specific knowledge can matter when it has not already been absorbed into a large model's training data. A recently published experiment on financial document analysis partly supports Mensch's point.
Bridgewater and Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, fine-tuned the open-source model Qwen3-235B using their own investor evaluations. According to their own assessment, the fine-tuned model reached 84.7 percent accuracy on financial documents, while the best frontier model reached 78.2 percent. Operating costs were nearly 14 times lower.
That result suggests a clear possibility: when an organization has internal expert knowledge that is not already in a general-purpose model, fine-tuning an open-source model can produce a useful advantage. It can also reduce operating costs in the example described.
But the caveats are important. The comparison was not independent, and both companies have a stake in selling their products. It was also only a snapshot. The source article notes that companies like Anthropic or OpenAI could buy that kind of data for future training or generate it themselves, which would likely put them back on top.
What companies should take from the debate
The strongest takeaway is not that every company must abandon proprietary AI models. The source does not support that conclusion. The clearer lesson is that AI adoption now requires a sharper view of data control, model control, and vendor dependence.
For businesses, the core questions are practical:
- Which data should remain inside open systems?
- Who decides what an AI model can access?
- Which workflows reveal valuable business processes?
- When does internal training data create a real advantage?
- How much dependence on a closed model provider is acceptable?
Mensch's argument is strongest when AI is close to a company's unique knowledge. In those cases, the risk is not just that an outside model performs a task. It is that the outside provider may gain insight into what makes the business work.
As companies move deeper into frontier AI, the question of ownership becomes more than a technical preference. It becomes a strategic decision about where growth, knowledge, and control will sit.