Microsoft and OpenAI on a Crusade for Open AI Models

Microsoft and Its Allies Defend Open AI Models

On July 24, an open letter titled Open Weights and American AI Leadership was initiated by Microsoft and signed by 235 influential companies in the field of artificial intelligence. Among these signatories are giants such as NVIDIA, Amazon, Y Combinator, The Linux Foundation, and more recently, OpenAI. This initiative aims to influence the U.S. government not to restrict the use of open-weight AI models, citing security concerns.

This letter is a direct response to fears that the U.S. government may consider bans or limitations on these models for national security reasons, a concern that has been highlighted by recent incidents involving models like Claude Fable 5. The signatories emphasize that closed models are not necessarily safer, as they can be vulnerable to compromises, abuses, or failures that are not easily detectable by external observers.

The Risks of Closed Models and the Case for Openness

Closed AI models, while they may seem secure, concentrate technological power in the hands of a few players, which can create single points of failure. This limits competition and entrusts crucial technology to a small number of providers. In contrast, open-weight models allow a diverse community of researchers and developers to scrutinize their functioning, identify potential flaws, and propose continuous improvements.

A notable aspect of the letter is its support for the technique of distillation, a process where models are trained from the outputs of other models. This method is essential for the development and evaluation of models, and it aligns with a tradition of learning and continuous improvement in technologies, similar to that of the open-source movement.

Distillation: A Technique in the Spotlight

Distillation is a common practice in the development of AI models, where the results of one model are used to train another. The signatories of the letter insist that policymakers should not confuse this legitimate technique with malicious practices. Distillation has been a driver of innovation, enabling the improvement and validation of existing models, and it is considered a cornerstone of technological progress in the field of AI.

Anthropic's Response and Its Concerns

However, not all companies share this enthusiasm for open-weight models. Anthropic, a company notably absent from the list of signatories, published its own response titled Our Position on Open Weights three days after the initial letter. CEO Dario Amodei expressed concerns about the possibility that authoritarian governments could develop AI models more powerful than those of the United States, which could be used for malicious purposes, such as cyberattacks or biological attacks.

Amodei also called for strict regulation of large-scale distillation operations, while clarifying that Anthropic does not support a total ban on open-weight models.

A Call for International Regulation

A few days later, on July 28, another letter titled Pacing the Frontier was published, signed by 1,324 employees from major AI companies, including figures such as Jakub Pachocki from OpenAI, Ilya Sutskever, and Dario Amodei from Anthropic. This letter calls on the U.S. government to support an international effort to develop the technical and governance tools necessary for thoughtful regulation of automated AI development.

The signatories express their concern over the increasing competitive pressure and the acceleration of AI advancements, exacerbated by automated research. With companies like Anthropic producing 80% of their code with Claude Code, and OpenAI reducing its service costs by 20%, the urgency to regulate these technological advancements is becoming increasingly pressing.