Editorial illustration for Nadella calls AI labs "ironic" for banning distillation while using others' data
Nadella Slams AI Labs for Banning Model Distillation
Satya Nadella has a problem with how his rivals write their terms of service. In a post on his personal blog, the Microsoft CEO singled out AI labs, OpenAI and Anthropic among them, for banning distillation, the technique where a smaller model learns by studying the outputs of a larger one. Both companies have used distillation bans to target Chinese AI firms accused of copying their models on the cheap.
Nadella's issue isn't the bans themselves. It's the double standard underneath them. These same labs train on public data under fair-use claims, restrict others from doing anything similar to their own models, and still mine customer interactions for further training. Nadella calls this contradiction "ironic," arguing it lets a handful of infrastructure operators capture the economic value that customers and their data actually generate.
The critique lands at a convenient moment for Microsoft, which sells the cloud infrastructure companies would need if they wanted to keep their own AI learning loops in-house rather than handing that value to a model provider. Nadella frames the stakes in blunter terms below.
The result, Nadella says, is that economic value concentrates with the infrastructure operators instead of the companies that actually generate the knowledge.
Why this matters
Nadella isn't wrong, but he's not neutral either. Microsoft has spent billions on OpenAI and competes directly with Anthropic in enterprise AI, so calling out the industry's fair-use-in, lockdown-out double standard also serves Microsoft's own interest in cheaper, more flexible model training. That doesn't make the point weaker.
The terms-of-service bans on distillation, aimed mostly at Chinese labs like DeepSeek, sit awkwardly next to years of scraping the open web under the same fair-use logic these companies now want to withhold from others. For developers and founders building on top of these platforms, the contradiction matters practically: it signals that "fair use" is being applied as a one-way door, open for incumbents, closed for challengers. Researchers should watch whether this kind of public pressure changes actual policy or just optics.
If Nadella's comments push OpenAI or Anthropic to clarify or loosen distillation terms, that's a real shift in how model training norms get set. If not, it's just one CEO needling his rivals while running the same playbook himself.
Related Reading
Policy & Regulation Grammarly faces class-action suit over AI ‘Expert Review’ feature
Policy & Regulation Claude Mythos highlights EU AI safety gaps, says researcher Caroli
Policy & Regulation After ditching AI fitness apps and a Fitbit, I return to Peloton classes
From Archives Researchers Propose Method to Block Illegal AI Content for Kids
From Archives Apple Sues OpenAI, Alleges Trade Secret Theft for Hardware
Common Questions Answered
Why does Satya Nadella consider AI labs' distillation bans ironic?
Nadella argues that OpenAI and Anthropic ban distillation—where smaller models learn from larger ones' outputs—while simultaneously training their own models on data scraped from others without permission. He views this as a hypocritical double standard where these companies restrict others' access to their knowledge while freely using external data sources for their own model development.
What is model distillation and why have AI labs banned it?
Model distillation is a technique where a smaller, more efficient model learns by studying the outputs of a larger model, effectively copying its knowledge at a lower cost. OpenAI and Anthropic have implemented distillation bans primarily to target Chinese AI firms like DeepSeek that are accused of using this method to create cheaper copies of their models.
According to Nadella, how does the distillation ban affect economic value distribution?
Nadella contends that by banning distillation while training on others' data, economic value becomes concentrated with infrastructure operators rather than with companies that actually generate the underlying knowledge. This creates an unfair advantage for well-resourced companies that control the infrastructure while preventing smaller competitors from accessing the same knowledge through legitimate technical means.
What is Microsoft's interest in challenging these distillation bans?
Microsoft has invested billions in OpenAI and competes directly with Anthropic in enterprise AI, so advocating against distillation bans serves its own commercial interests in accessing cheaper and more flexible model training options. However, the article notes that while Nadella's criticism may be motivated by self-interest, it still highlights a legitimate inconsistency in the industry's fair-use policies.
Further Reading
- Microsoft CEO Nadella subtly criticizes dual standards among AI labs on distillation - Hyper.ai
- Satya Nadella's AI Warning Is a Sales Pitch - ProMarket
- Microsoft's Satya Nadella: We Can't Let AI Giants Eat the Economy - The Wall Street Journal
- Distillation: The New U.S.–China AI Fight - Forbes
- Detecting and preventing distillation attacks - Anthropic
