Дарио Амодеи из Anthropic: не против open-weight, но опасается китайского ИИ

Anthropic founder and CEO Dario Amodei publicly clarified his position on open-weight models on July 27, 2026: he is not opposed to them in principle, but names the rapid growth of Chinese AI capabilities as the main risk. This is reported by TechCrunch.

What Exactly Amodei Said

Dario Amodei separated two issues that are usually conflated in public discussion: the openness of model weights and geopolitical competition with China. According to TechCrunch, the Anthropic chief does not consider the publication of weights itself a threat — what worries him is the pace at which Chinese labs are closing the gap with Western ones.

For Amodei, the debate over "openness versus closedness" is secondary to the question of who sets the technological pace in the industry. This shifts the focus: not "whether to allow open models," but "who releases ever more powerful AI faster."

What Open-Weight Models Are

Open-weight models are ones whose developer publishes trained weights for free download, fine-tuning, and deployment on one's own hardware. Meta (the Llama family), France's Mistral, and China's DeepSeek follow this approach on the market. Anthropic and OpenAI, by contrast, keep the weights of their flagship models closed and provide access only via API.

Anthropic, founded in 2021, has never released the weights of its Claude model to the public. That is why its public stance on open-weight is seen as an indicator of sentiment across the entire closed segment of the industry.

*The statement was made on July 27, 2026, source: TechCrunch

*Amodei is not opposed to open-weight models as a class

*The main concern is the speed at which Chinese AI is growing

*Anthropic (founded in 2021) traditionally does not publish Claude's weights

Why It Matters

Amodei's position matters because Anthropic is one of the three leading Western AI labs, alongside OpenAI and Google DeepMind, and its founder has long taken part in debates over how the US should regulate access to advanced AI technology. The distinction "open weights are fine, but the pace of Chinese AI is not" shifts the emphasis of the discussion: from banning model publication to preserving technological leadership.

For regulators and developers, these are different agendas. A ban on open weights hits researchers and startups within the country; the race with China is already about export controls, funding, and the speed of model releases.

The Context of the Open-Weight Debate

The debate over open weights has been going on in the industry for several years now. Advocates of openness argue that publishing weights speeds up research, reduces dependence on a handful of vendors, and makes the technology verifiable. Opponents point to the risk: once weights are published, they cannot be recalled, and they can be fine-tuned for undesirable purposes.

Amodei's position here is atypical: representing a closed lab, he does not fully side with banning open models, but instead directs the concern toward a specific external factor — the growth of Chinese AI.

As

TechCrunch reports, the essence of Amodei's position is that the problem is not the open weights themselves, but the speed at which Chinese AI is closing the gap. — as reported by TechCrunch, July 27, 2026

What This Means

The head of Anthropic indicated that the dividing line in the industry runs not along "open versus closed models," but along the line of geopolitical competition. If this framing takes hold, the next wave of debate and regulation in the US will be less about banning open-weight and more about how to maintain the lead over Chinese labs. So far, Anthropic has not backed up the statement by publishing its own open-weight models — the company remains in the closed camp, and its signal is addressed more to policymakers and the industry than to developers of open models.

*Meta has been designated an extremist organization and is banned in Russia.