Три открытых письма за пять дней: Microsoft, Anthropic и 1324 разработчика об open-weight

In late July 2026, the AI industry exchanged three open letters about the future of AI development: 235 companies led by Microsoft demanded protection for open models, Anthropic responded with its own position, and 1,324 employees of leading AI labs called on the US to support an international slowdown of the race. All of this happened within five days.

What the open-weights letter demands

The letter "Open Weights and American AI Leadership", organized by Microsoft and published on July 24, 2026, was signed by 235 companies, including NVIDIA, Amazon, Y Combinator, The Linux Foundation, and (later joining) OpenAI. It is directed against possible attempts by US authorities to ban or restrict open-weight models under the pretext of "safety" concerns — a risk the industry began discussing after a series of restrictions on large AI systems.

• Publication date — July 24, 2026
• Number of signatories — 235 companies
• Among the signatories: NVIDIA, Amazon, Y Combinator, Linux Foundation, OpenAI
• Main thesis: closed models are no less vulnerable than open ones, and their monopolization is more dangerous

"Relying exclusively on closed models is itself unsafe: they can be hacked, misused, or create situations that observers cannot detect.

Concentrating AI capabilities in the hands of a small number of providers creates single points of failure and weakens competition," states the letter organized by Microsoft.

A surprising point was the support for distillation — the practice whereby one model is trained on the output of another. The authors called it "a legitimate model improvement technique" with a long history in open-source software traditions and warned regulators not to confuse distillation with copyright infringement.

Why Anthropic refused to sign

Anthropic did not join the 235 signatories and on July 27, 2026 published its own position on open-weight models. CEO Dario Amodei hardened the company's rhetoric: he named authoritarian governments capable of building "models more powerful than American ones" and the misuse of models for "cyberattacks or biological attacks" as the main threats.

Anthropic called for curbing "industrial-scale distillation operations" — a message directly opposed to the Microsoft letter. Nevertheless, Amodei noted: "Anthropic has never called for a ban on open-weight models" — making clear that the company is not advocating a total ban, but rather selective restrictions.

Who called for slowing down the AI race

On July 28, 2026, the letter "Pacing the Frontier" was published, signed by 1,324 employees of leading AI companies. Among the authors are Jakub Pachocki (Chief Scientist at OpenAI), Ilya Sutskever (Safe Superintelligence Inc), Dario Amodei and Jack Clark (Anthropic).

"We ask the US government to support international efforts to develop

the technical and governance tools necessary for the intentional moderation of the pace of automated AI development," reads the "Pacing the Frontier" letter.

The authors are concerned that automated AI research is accelerating progress faster than the industry can develop safety mechanisms. The examples are telling: according to Anthropic, the company generates 80% of its code using Claude Code; OpenAI reduced end-to-end service costs by 20% using the Sol model; Kimi K3 designed its own chip for a nano-model based on its own architecture. In other words, AI is already actively creating the next generation of AI.

What this means

Three letters in five days reveal a deep rift in the industry: some are betting on openness and competition, others are demanding international braking, Anthropic is choosing selective regulation. The regulatory season for AI in 2026 has entered a hot phase, and the stakes are high — the question is who will control, and how, the most influential technology of the decade.

Frequently Asked Questions

What are open-weight AI models?

Open-weight models are AI systems with publicly available parameters: they can be downloaded, studied, run locally, and fine-tuned. Unlike closed models (GPT-5.6, Claude), their behavior is transparent to independent researchers, which, in the opinion of Microsoft and NVIDIA, is itself a protection against misuse.

How does

Anthropic's position differ from the Microsoft letter?

Microsoft with 234 co-signatories considers open-weight models a less monopolistic and more transparent risk than closed ones. Anthropic is not generally opposed to open models, but demands restrictions on industrial distillation and advocates international regulation of the pace of AI development — a position aligned with "Pacing the Frontier".