July 20, 2026, 3:40 pm | Read time: 3 minutes
Artificial intelligence has become a part of everyday work for many companies. Employees create texts, analyze data, or automate processes using AI systems. In doing so, they often disclose internal information, business processes, or other company data to the applications. Microsoft CEO Satya Nadella sees this as a problem.
In a recent blog post, he warns that companies not only pay for the use of AI systems but could also inadvertently share valuable knowledge with their providers.
AI Learns from Daily Use
According to Nadella, the real danger does not lie solely in the use of AI but in the data that continuously flows into the systems. The more frequently companies work with a model, the more knowledge is embedded into the respective AI, he says.
It’s not just the inputs from users that matter. Corrections and feedback also contribute to the systems’ learning. Nadella describes this connection as follows: “Models learn from ‘byproducts’—the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong. Every correction becomes part of the institutional knowledge.”
This means companies are disclosing information that competitors would find difficult to obtain otherwise. Nadella is particularly critical of contract terms that allow providers to learn from usage data and interactions.
More Control Over Data and Systems
For this reason, the Microsoft CEO calls for a different approach to AI applications. He believes companies should retain control over their data and not become permanently dependent on a single provider.
Nadella suggests so-called orchestration layers as a possible approach. These would allow companies to switch flexibly between different AI models, preventing them from being tied to a single platform in the long term.
Additionally, Nadella advocates for proprietary learning environments in the cloud. There, companies could manage and control their data, inputs, and feedback themselves. His goal is to strengthen companies’ data sovereignty.
AI Costs Skyrocket: Company Spends $500 Million in Just One Month
New Microsoft AI Aims to Hook Users
Open Models Become More Attractive to Companies
Solo.io also reports that many companies are now seeking alternatives to proprietary AI models. The U.S. company develops network and security software for managing AI systems in businesses.
According to founder and CEO Idit Levine, many customers, after initial experiences with proprietary models, are increasingly interested in other solutions. Instead of using external AI services, they are turning to open models that can be operated locally. This allows companies to maintain control over their data while also reducing costs.
Also of interest: New Microsoft AI aims to make users addicted
According to the company, demand for such models is rising. For many firms, control over their own data is becoming as important as the performance of the AI they use.
Nadella’s Appeal Carries Special Weight
Nadella’s warning is particularly noteworthy in light of Microsoft’s investments. The company has invested billions in both OpenAI and Anthropic. Nevertheless, the CEO urges companies not to relinquish their data and knowledge lightly.
Nadella sums up his central message in two sentences: “By using intelligence, you create intelligence. And what you create should belong to you.”