Microsoft CEO Slams AI Labs Over Distillation

Satya Nadella, Microsoft’s CEO, calls out AI labs for hypocrisy regarding distillation criticism, a remark widely interpreted as targeting Anthropic. Credit: Getty Images

Microsoft CEO Satya Nadella challenges AI developers for limiting model distillation while simultaneously leveraging public and customer data to train their own systems.

Microsoft’s Satya Nadella has publicly criticized unnamed AI labs across the tech industry for their contradictory stance on AI distillation, a technique where a smaller model is trained using the outputs of a larger, more powerful system.

In a post on X, Nadella argues that model creators complaining about distillation are being hypocritical about the broader AI training process.

He acknowledges that ‘the great innovation that comes from model providers having fair use rights to train on public data is needed’, yet finds it ‘ironic’ that the industry consensus imposes ‘restrictive terms’ on distillation.

This occurs even as these companies reserve the right to learn from customer usage and interaction data themselves.

“If learning flows in only one direction, economic value converges toward the owners of the learning infrastructure rather than the creators of the knowledge itself,” Nadella notes.

Anthropic attacks highlight growing sourcing row

Anthropic, OpenAI, and Google DeepMind rely on third-party content to train their AI models. ChatGPT, Claude, and Gemini source information from publicly available writing, images, and data.

Many have raised ethical and copyright concerns about this training method, leading to legal action from major content creators.

For instance, BBC News reported that the BBC is suing Anthropic for non-consensual data sourcing.

Although Nadella did not name Anthropic directly, evidence suggests the post was strongly directed at the AI firm. His criticism follows an earlier Anthropic post on ‘detecting and preventing distillation attacks’.

In a June letter to South Carolina Senator Tim Scott and Massachusetts Senator Elizabeth Warren, Anthropic claimed Alibaba recently executed the ‘largest known distillation attack’ against its data.

The statement argued that competitors could use distillation to acquire ‘powerful abilities from other labs in a fraction of the time and at a fraction of the cost, that it would take to develop them independently’. Alibaba has yet to respond to these accusations.

Companies like the BBC and Reddit have sued Anthropic, accusing it of using their data to train its Claude model without consent (Credit: Anthropic)

Model outsourcing raises critical business risks

In the same X post, Nadella warns that companies relying on major AI models are granting AI firms access to their proprietary data.

He emphasizes that companies must own their AI infrastructure and industry knowledge rather than depending on a single model vendor.

Nadella advises that firms should conduct their own evaluations and ‘learning loop’ to ensure their AI technology develops and improves over time.

Enterprises require strict security parameters to protect proprietary developments. Nadella states that businesses need a real trust boundary for their human and token capital to compound.

He adds that this boundary must not be crossed by anyone or anything, ‘not even the intelligence exhaust, without consent’.

Similarly, Elon Musk has publicly criticized Anthropic’s data sourcing practices for training its AI models.

In a February post on X, Musk accused Anthropic of ‘stealing training data at a massive scale’, predicting it would face ‘multi-billion dollar settlements’ for theft.

“This is just a fact,” he adds.