Mistral founder Arthur Mensch is making the case for open-source AI. In a LinkedIn post, he warns companies against depending on closed AI models.

Companies that sell closed models are storing more and more data, giving them a window into their customers' business processes, Mensch claims. Some AI labs "have a track record of going after their most successful customers thanks to this information," according to Mensch.

He advises companies to store their data in open systems, set 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," Mensch writes.

Mensch's comments follow similar remarks by Palantir CEO Alex Karp, who also urged companies to build their own AI models instead of relying on proprietary outside solutions. Palantir also published a manifesto for secure AI in business. Among other things, it reads, "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."

Mensch has a point, but he also has a business to run

Mensch's arguments are valid, but they need context. Mistral is the only EU company with relevant AI models, and it can't really compete with top-tier models like GPT-5.6 Sol or Fable 5 on raw performance. Mistral's business model leans heavily on EU sovereignty because that's where the company stands to gain the most, even though about 30 percent of its shares are held by US investors. Large general-purpose AI models have also repeatedly beaten specialized models on specialized benchmarks, as long as the relevant domain knowledge was part of the training data. Mensch is arguing his own book here.

A recently published experiment on financial document analysis partly backs him up, though. Internal expert knowledge that wasn't included in the training data of large models can provide an edge.

The hedge fund 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 hit 84.7 percent accuracy on financial documents, while the best frontier model reached 78.2 percent. Operating costs were nearly 14 times lower.

That wasn't an independent comparison, and both companies have a stake in selling their products. It's also just a snapshot. Companies like Anthropic or OpenAI could simply buy that kind of data for future training or generate it themselves, which would likely put them back on top.