Microsoft CEO Nadella warns: AI models are Trojan horses for enterprise data

Satya Nadella, CEO of Microsoft, issued a stark warning in a recent blog post: enterprises using proprietary AI models from labs like OpenAI and Anthropic are unknowingly surrendering their most sensitive business data. He argues that companies 'pay for intelligence twice' — first with money for token usage, and second with 'something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful.' Every prompt, every correction to a model's output, and every agent action becomes 'exhaust' that model makers can distill into institutional know-how. 'The better you want the model to perform, the more of that knowledge you have to feed it,' Nadella writes. He calls out the hypocrisy of AI labs training freely on public internet data while restricting others from doing the same to their models through restrictive distillation terms. This echoes concerns from VCs and Palantir CEO Alex Karp about AI labs acting as 'Trojan horses' that could eventually compete with their own customers.

Nadella's proposed solution is predictably cloud-centric: enterprises should 'retain ownership' of all interaction data — prompts, feedback, usage patterns — and build 'proprietary learning environments' on cloud platforms (implicitly Microsoft Azure). He also advocates for 'orchestration layers' that allow easy switching between multiple AI models, avoiding vendor lock-in — a trend already accelerating via tools like AI gateways. The unspoken subtext is open-source. Idit Levine, CEO of Solo.io (whose technology powers the Linux Foundation's Agentgateway project), confirms her enterprise customers — including T-Mobile, ADP, and SAP — are increasingly moving to on-premise open-source models after experimenting with proprietary ones. 'Can I take an open source model and run it on-prem? It will do almost 90% of what the big one's doing. It will cost way less,' she explains. This shift gives companies control over their data while cutting costs and limiting exposure to model makers.