Satya Nadella criticises AI firms over model distillation stance

Microsoft CEO Satya Nadella has appeared to criticise leading artificial intelligence companies, including Anthropic, over their stance on AI model distillation, calling it "ironic" that firms benefiting from publicly available data oppose similar practices by others.

In a post on X, Nadella argued that AI companies should not restrict model distillation while continuing to learn from customer interactions and usage data.

Nadella questions AI model distillation complaints

Without naming any company directly, Nadella said it was contradictory for AI developers to rely on public data to train their models while objecting to others using model distillation to build competing systems.

Model distillation is a technique in which a smaller AI model learns from the outputs of a more advanced model, allowing developers to create capable systems with fewer resources.

Nadella said that if knowledge flows in only one direction, infrastructure owners benefit financially while creators of knowledge are left behind.

Remarks appear aimed at Anthropic

Although Nadella did not mention Anthropic by name, his comments came months after the AI company publicly raised concerns about model distillation.

Earlier this year, Anthropic CEO Dario Amodei alleged that Chinese AI developers were using Claude to train rival models. The company later claimed Alibaba had carried out what it described as the largest known distillation attack against its systems.

Anthropic has argued that model distillation enables competitors to replicate advanced AI capabilities more quickly and at a lower cost than developing them independently.

Alibaba has not publicly responded to those allegations.

Nadella urges businesses to own AI infrastructure

Nadella also warned companies against becoming overly dependent on third-party AI models.

He said businesses should retain ownership of their AI infrastructure, evaluation processes and institutional knowledge rather than relying on a single model provider.

According to Nadella, organisations need clear boundaries to protect proprietary data and ensure that information is not shared without consent.

His comments add to the ongoing debate over AI training practices, data ownership and intellectual property as competition among major AI developers intensifies.