Anthropic chief executive Dario Amodei took to his blog Monday afternoon to put the record straight after industry chatter suggested his company backs a U.S. ban on open‑weight AI models, particularly those from China. "Anthropic has never advocated for a ban on open‑weights models," he wrote, emphasizing that his past statements make his stance clear.
The clarification came on the heels of Nvidia founder Jensen Huang’s open letter on X, signed by dozens of AI firms including Hugging Face, Meta, Microsoft, and Mistral. The letter warned policymakers against imposing "premature restrictions" on open‑weight models, but it did not name China. The broader debate, however, has focused on allegations that Chinese AI labs are advancing by appropriating intellectual property from U.S. counterparts, often through a technique called distillation—prompting a model repeatedly to extract its inner workings.
Amodei distinguished between the technology itself and the geopolitical threat it poses. He described open‑weight models that lack dangerous capabilities as a public good, noting that they cost nothing beyond the compute needed to run them and that they deliver value to businesses, developers, and researchers. "Businesses using open‑weight models, even those from China, are not among the risks I fear," he said.
The CEO’s primary concern, he explained, is that authoritarian governments could develop AI systems more powerful than those in the United States, using them to secure permanent military superiority or to suppress their own populations. While he named the Chinese Communist Party as the most capable of these regimes, he stressed that the threat is not limited to China.
Amodei also warned that open‑weight models could become especially hazardous in the context of biological attacks. Because such models are released without a central authority, applying guardrails or monitoring usage is difficult. He cited a UK AI Security Institute report indicating that once open‑weight models are released, they cannot be withdrawn.
To counter the Chinese threat, Amodei outlined three concrete actions. First, he called for continued restrictions on China’s access to advanced chips—a policy the United States has pursued for years. Second, he advocated for a formal crackdown on distillation practices that facilitate intellectual‑property theft. Third, he expressed support for a global model‑safety testing organization that would assess the most capable systems regardless of origin, while exempting less powerful models from startups and academia.
The CEO praised recent moves by the Trump administration and industry proposals aimed at universal testing standards. He argued that effectiveness would require participation from the CCP, noting that limited cooperation on preventing AI‑enabled biological weapons might be possible because it aligns with China’s own interests.
Amodei’s statements arrive as Washington weighs further sanctions against China for alleged IP theft involving U.S. AI models. By separating the debate over open‑weight accessibility from the broader geopolitical risk, he hopes to keep the innovation ecosystem vibrant while addressing the most pressing security concerns.
In his view, a balanced approach—preserving open‑weight benefits while tightening controls on the most dangerous applications—offers the best path forward for the industry and for global stability.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.