Are businesses training their rivals?

Microsoft CEO Satya Nadella has warned that companies using proprietary AI models may be giving model providers access to valuable business information.

Businesses do not just pay companies such as OpenAI and Anthropic to use their models.

They also share internal data, prompts, feedback and corrections to make the AI more useful.

Nadella argues that companies are effectively paying twice: once with money and again with their knowledge.

Every prompt and correction can reveal how a business works, including its processes, tools and internal expertise.

Critics fear AI providers could use this information to build products that compete with their own customers.

Nadella also criticised restrictions on AI “distillation”, where developers study a model’s outputs to train a smaller or cheaper model.

He argues that AI companies should not be allowed to train on public internet data while stopping others from learning from their models.

His solution is for businesses to keep control of their prompts, feedback and data by building secure AI systems in the cloud.

He also recommends using tools that let companies switch between different AI models instead of relying on one provider.

This is already pushing more businesses towards open-source models, which can be cheaper and run on a company’s own systems.

Solo.io CEO Idit Levine said some open models can deliver around 90% of the performance of leading proprietary models while giving companies more control.

Vercel and OpenRouter are also seeing more demand for open models.

Last month, they made up 29% of traffic through Vercel’s AI gateway.

Nadella’s wider message is simple: companies should think about what AI providers may learn from their usage, not just what the technology costs.

The intern signed an NDA. The AI model got an API key and the entire company wiki. - MG