Satya Nadella calls out AI's model-cloning double standard
Business Insider ● Covered by 2 sources
Satya Nadella just called out AI labs for training on everyone else's data, then crying foul when others do it to them. He says it's hypocrisy — and enterprises are paying the price for it.
Based on reporting by Business Insider — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Satya Nadella didn't name names, but the target of his Sunday X post was pretty obvious. The Microsoft CEO took aim at model makers, Anthropic chief among the unnamed suspects, who scrape the open web to train their systems and then throw up walls the moment someone tries to learn from their outputs in return.
His argument boils down to a simple asymmetry. Companies like Anthropic, OpenAI, and Google DeepMind built their models on public writing, images, and data scraped from across the internet, much of it without consent, which is why there's a pile of lawsuits over nonconsensual scraping working through the courts right now. But once those models exist, the same companies impose tight restrictions on distillation, the technique of training a smaller model off a bigger one's outputs, while quietly reserving the right to keep learning from how customers use their products. Nadella called that setup ironic. Business Insider's Alistair Barr, digging into the same story, called it delicious irony, which might be the more honest description.
The backdrop here is Anthropic's ongoing fight with Chinese competitors. CEO Dario Amodei has accused Chinese labs of stealing Claude's work to train their own models, and last month Anthropic told senators Tim Scott and Elizabeth Warren that Alibaba pulled off the largest known distillation attack the company has seen. Anthropic's February statement warned that distillation lets rivals grab powerful capabilities in a fraction of the time and cost it takes to build them from scratch. Alibaba never responded publicly. Elon Musk piled on too, accusing Anthropic of stealing training data at massive scale and paying settlements over it.
What Nadella's really building toward, though, isn't a gotcha about Anthropic specifically. It's a pitch for why companies shouldn't lean on any single model vendor at all. He argues enterprises relying entirely on a leading model are effectively handing over their proprietary data and paying for the privilege. His fix is that businesses should own their own AI infrastructure and institutional knowledge, run their own evaluations, and build a learning loop that keeps improving on its own terms. He described it as a hard boundary that nothing crosses without consent, not even what he called intelligence exhaust.
Anthropic didn't respond to a request for comment.
My take — AI-written commentary, not fact-checked reporting
Nadella's not wrong about the hypocrisy, but he runs Microsoft, which has its own massive stake in how AI training data gets treated, so this isn't some neutral referee call. The real tell is that his solution conveniently points enterprises toward owning their own infrastructure rather than trusting any single vendor's model — which happens to be a pitch Microsoft is well positioned to sell. Everyone in this fight wants rules that favor the direction their own data flows.
Read more about this at: Business Insider