The struggle over open source: Tech giants align against the White House

More than 50 companies and organizations, including Nvidia, Microsoft, Google, Meta, and OpenAI, have published an unusual letter calling on the US administration not to impose sweeping restrictions on open AI models. And why did Anthropic stay out of the move?

The debate over the future of artificial intelligence reached a boiling point this past weekend. More than 50 companies and organizations, including Nvidia, Microsoft, Google, Meta, OpenAI, AMD, IBM, Cisco, Palantir, Mistral, and many others, signed an open letter calling on the US administration not to impose preemptive restrictions on open artificial intelligence models, which would essentially undermine the development and learning capabilities of models known as Open Weight.

The letter was published in the midst of an ever-widening confrontation between the artificial intelligence industry and Washington. In recent weeks, the discussion has intensified around the rapid progress of Chinese companies, led by Moonshot AI, which developed Kimi K3, one of the most advanced Open Weight models published to date.

At the same time, senior officials in the US administration have accused the company of stealing intellectual property through distillation, a method in which a new model is trained based on the output of a more advanced model. Against this backdrop, the companies are seeking to send a clear message: while the law against intellectual property theft must be enforced, sweeping restrictions on the development and use of open models should not be imposed.

Open models are necessary

The letter, published by Nvidia CEO Jensen Huang in his first-ever post on the X network, marks a significant shift in the discourse surrounding artificial intelligence. While for years the industry was divided between companies that supported open models and companies that chose to keep their models behind commercial APIs, today, according to the letter, most major players agree on one principle: open models are an integral part of the United States' ability to continue leading the field of artificial intelligence.

According to the signatories, the American advantage is measured not only by its ability to develop the most powerful model in the world, but also by its ability to build a broad ecosystem that will allow researchers, companies, and governments to adopt the technology and develop it. The letter argues that open models contribute to innovation, strengthen cybersecurity through broad community review, and allow countries and organizations to maintain technological sovereignty instead of being dependent on a limited number of suppliers.

However, it states that violations of the law or theft of intellectual property should not be ignored. On the contrary, they write that such cases should be handled through targeted legal and commercial tools, rather than through sweeping restrictions on technology that serves thousands of companies and developers.

Fear of Chinese distillation

As mentioned, the timing is not coincidental. The same debate between the administration and the industry is sharpening against the backdrop of the rapid progress of Chinese AI companies. Moonshot AI's Kimi K3 has become one of the most talked-about open models in the world in recent weeks, and in the US, it is claimed that some of its capabilities were achieved through the distillation of American models.

Thus, Michael Kratsios, Director of the Office of Science and Technology Policy at the White House, publicly accused Moonshot AI of developing the model using the output of Fable 5, Anthropic's leading model. Treasury Secretary Scott Bessent also said that the administration is considering measures against companies that steal intellectual property from American companies, and even hinted at the possibility of sanctions.

This is exactly where the gap between the administration and the technology companies is created. While Washington is considering tightening oversight of open models following concerns about intellectual property theft by Chinese companies, most companies warn that broad restrictions could primarily harm the American industry.

Where is Anthropic?

One of the most intriguing aspects of the letter is the identity of the signatories. Alongside companies that have been associated with open source for years, such as Mozilla, Hugging Face, and the Linux Foundation, there are also companies that develop commercial and closed models, including Microsoft, Google, Meta, and OpenAI. OpenAI itself was not among the first signatories, but joined later. The company's CEO, Sam Altman, also expressed public support for the message that the United States should lead in both open and proprietary models. However, according to reports in the US, OpenAI continues to promote measures before the administration that would increase oversight of models that it claims were trained using illegal distillation.

In the broad list of companies, Anthropic's absence is particularly notable. The company, which develops the Claude and Fable model series, does not appear among the signatories, and Amazon, which is also one of the key investors in Anthropic, did not join the signatories either. Neither of the two has published an official explanation for its decision.

Anthropic has for some time taken a more cautious approach regarding advanced AI models. The company warns that once model weights are published to the public — the files on which the model is based — anyone can copy, modify, and distribute them, and therefore it is almost impossible to control how they will be used. At the same time, Anthropic was among the first companies to warn against illegal use of distillation by Chinese companies.

The letter presents a rare front of dozens of companies and organizations that have stood behind a common position: intellectual property violations must be enforced, but sweeping restrictions should not be imposed on open models. Now it remains to be seen whether this message will influence the policy that Washington will formulate.