SAN FRANCISCO — Anthropic CEO Dario Amodei said the company does not advocate for banning open-weight artificial intelligence models, addressing criticism over its decision not to sign an industry letter supporting open-source AI development.

The letter, which circulated among AI industry leaders last week, called for greater transparency and accessibility in the development of advanced AI systems. Anthropic's absence drew pointed questions about where the company stands.

The answer matters commercially. Anthropic spent roughly $800 million on compute costs over the past 12 months training its Claude models—a capital outlay that only makes sense if the company can defend a proprietary position. At an $18.4 billion valuation following a Google-led $2 billion funding round, the market is pricing in exactly that: premium enterprise contracts, controlled deployment and durable intellectual property that open-weight competitors cannot easily replicate.

Open-weight models, which expose underlying code and parameters for public use, lower barriers to entry for smaller developers and research institutions. Proponents argue they speed iteration cycles and broaden the developer ecosystem. Critics inside companies like Anthropic counter that proprietary control enables more rigorous safety testing and limits potential misuse of powerful capabilities.

The commercial logic aligns Anthropic with OpenAI, which also holds its foundational models close. It puts both companies in direct contrast with Meta, whose Llama family is released as open-weight—a strategy that trades monetization upside for ecosystem influence and developer adoption.

Policymakers in the United States and Europe are watching these choices closely as they draft frameworks to govern advanced AI, with transparency and access central to both debates.

Amodei's clarification stops well short of endorsing open-weight development. What the company opposes is a ban—not open-weight models themselves. For a business burning $800 million a year on compute to stay ahead, that distinction is the entire strategy.