Microsoft and Nadella Call Out Anthropic's Hypocrisy

Satya Nadella Takes Aim at AI Manufacturers' Practices

Satya Nadella, the CEO of Microsoft, recently expressed his dissatisfaction with certain practices of artificial intelligence model manufacturers, particularly Anthropic. In a post on X, Nadella labeled the complaints from these companies regarding distillation attacks as hypocritical. Model distillation is a technique that involves training an AI using the results of a more advanced model. According to Nadella, this method jeopardizes the ownership and control of data by companies that rely on these models.

Model Distillation Under Fire

Model distillation is a process where a less powerful AI model is trained using the results of a more capable model. Nadella emphasized that while innovation stemming from the fair use of public data is essential, it is paradoxical for companies to impose restrictions on distillation while reserving the right to learn from user interactions and data. He added that if learning is one-sided, only the owners of the learning infrastructures financially benefit, leaving content creators behind.

AI Giants and the Use of Public Data

Large companies like Anthropic, OpenAI, and Google DeepMind rely on public data to train their AI models. Tools like ChatGPT, Claude, and Gemini acquire their intelligence from texts, images, and other publicly accessible data. This practice has led to lawsuits from companies and individuals accusing these AI labs of "scraping" content without consent.

Anthropic in Nadella's Crosshairs

Although Nadella did not explicitly name Anthropic, his comments seem particularly directed at this company. Earlier this year, Dario Amodei, CEO of Anthropic, complained that Chinese model manufacturers were using Claude to train their own models, accusing these companies of stealing their work. In September, Anthropic sent a letter to U.S. Senators Tim Scott and Elizabeth Warren, claiming that Alibaba had conducted "the largest known distillation attack" against them.

Anthropic's Accusations Against Alibaba

Anthropic stated that competitors could use distillation to quickly and cheaply acquire capabilities from other labs. This statement was made in February, but Alibaba has not publicly responded to the accusations. Nadella warned that companies relying on leading models risk losing control of their proprietary data and may have to pay to use it.

Nadella's Call for Business Autonomy

Nadella stressed the importance for companies to own their own AI infrastructure and develop their institutional knowledge rather than depend on a single model provider. He encouraged companies to conduct their own assessments and create an internal "learning loop" to continuously improve their AI capabilities. According to him, it is crucial to establish a true "trust frontier" to protect the human and intellectual capital of companies, ensuring that nothing crosses this frontier without consent.

Elon Musk Joins the Criticism

Elon Musk also criticized Anthropic, accusing it of stealing training data on a large scale and having to pay costly settlements for these practices. In a post on X in February, Musk claimed that these accusations were well-founded. Anthropic did not immediately respond to requests for comment from Business Insider.