Anthropic’s co‑founder Dario Amodei broke a week‑long silence on Sunday to clarify the company’s position regarding open‑weight AI models. In a detailed post, he said Anthropic has never advocated for a blanket ban on such models, describing those that lack dangerous capabilities as a "public good." The clarification follows an open letter signed by Nvidia CEO Jensen Huang and dozens of tech firms, as well as media reports that U.S. officials are considering prohibitions on Chinese open‑weight models.
Amodei outlined three policy measures he supports instead of a ban. First, he calls for tighter chip export controls and a crackdown on the "rampant smuggling" of advanced semiconductors to China. He argues that without access to U.S. chips, Chinese labs would struggle to train models that rival American ones under existing scaling laws. Second, he urges action against industrial‑scale model distillation, a process that lets competitors reproduce powerful models at a fraction of the original training cost. Third, he proposes mandatory safety testing for any model deemed sufficiently capable, whether open or closed, while exempting less capable research from startups and academia.
The open‑letter supporters argue that open weights broaden access and competition. Amodei agreed that broader availability can foster innovation, but he warned that open weights do not automatically favor defenders over attackers. He cited the potential for a capable model to weaponize a pandemic‑level virus using readily available materials, noting that defensive measures against such threats would take years to develop.
Anthropic’s most pointed complaint centers on model distillation. The company previously accused Alibaba’s Qwen lab of running the largest distillation campaign against Anthropic’s Claude model, allegedly using 25,000 fake accounts to generate 29 million exchanges. Amodei acknowledged that identifying and halting distillation is challenging because the offending accounts often emerge only after substantial copying has occurred. He stressed that policy intervention, rather than reliance on any single company’s enforcement, is needed to address the issue.
In the broader context, the clarification arrives as Washington debates how to regulate AI advances that could have national‑security implications. Reports suggest that U.S. officials are weighing restrictions on Chinese open‑weight models, a move that could reshape the competitive landscape for AI development. Amodei’s proposed measures aim to strike a balance: preserving the benefits of open research while mitigating risks tied to illicit chip procurement, unchecked distillation, and untested powerful models.
Anthropic’s statement underscores a nuanced position: the company does not oppose open‑weight models per se, but it rejects the notion of allowing them to proliferate unchecked, especially when they stem from stolen data, smuggled hardware, or lack rigorous safety evaluation. Whether policymakers will adopt the specific safeguards Amodei recommends remains to be seen, but the company’s public stance signals a willingness to engage in the ongoing policy debate.
Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.