Microsoft CEO Nadella warns enterprises about AI labs mining their business data

TechCrunch reports that Microsoft chief executive Satya Nadella published a blog post on a Sunday warning enterprises that the AI labs selling proprietary models can gain access to their most sensitive business information. The article frames this as a widespread worry in Silicon Valley, voiced by figures ranging from investor Jason Calacanis to Palantir chief executive Alex Karp, that labs could use what they learn to compete with their own customers.

Nadella argues that buyers effectively pay twice: once in money for token usage, and again through the proprietary knowledge they must reveal to make the models useful. In his account, models learn from the "exhaust" of usage, including prompts, the tools agents use and the corrections people make, and each correction becomes institutional know-how that a competitor could not purchase. He also contends it is inconsistent for labs to rely on fair use rights to train on public data while imposing restrictive terms on distillation, meaning the use of a model's outputs to train another model. He is especially troubled when providers reserve the right to learn from customer interaction data.

His proposed remedy is for companies to keep ownership of their data, including prompts and feedback, by building their own proprietary learning environments in the cloud and adding orchestration layers that make it easy to switch between model providers. TechCrunch notes that this advice suits a large cloud provider and that open source is an evident subtext.

The article adds supporting signals: Solo.io chief executive Idit Levine says customers increasingly ask about running open source models on their own premises, and open models made up 29% of traffic through Vercel's gateway last month. OpenRouter is also reported to see growing traffic to open source models.