Dario Amodei, the head of Anthropic, recently took to TechCrunch to clarify a growing narrative in the AI industry: that his company is somehow opposed to open-weight models. He doesn’t. In fact, he explicitly states that Anthropic has never advocated banning such models. That’s a crucial distinction, because the conversation around open-weight AI, models whose weights are publicly available, has become entangled with geopolitical tensions, particularly around China’s role in AI development.
Amodei frames open-weight models as a public good, especially when they lack dangerous capabilities. He’s not against transparency or open science, he’s just wary of how those models are being developed and deployed. His concern isn’t with the openness itself, but with the methods being used to scale and refine models in Chinese labs, particularly through techniques like model distillation. Distillation, in this context, means taking a large, powerful model and compressing it into a smaller, more efficient version, often without full disclosure of the original’s training or architecture. This process, Amodei suggests, may be accelerating the development of advanced AI capabilities without the same level of oversight or accountability.
He’s not saying open-weight models from China are inherently risky, in his view, they pose no direct business threat. But he’s concerned about the pace and opacity of development. “If you’re building something that’s going to be used in the real world, you want to know how it was built,” he said. “And if you’re not sure, you’re not sure you’re safe.”
This isn’t just a technical concern, it’s a strategic one. The U.S. government has been increasingly focused on restricting Chinese access to advanced AI tools and technologies, and Amodei’s comments come amid speculation that Anthropic may be involved in those efforts. He’s not denying that the U.S. is taking a hard line on Chinese AI, but he’s also not aligning with the idea that open-weight models are the problem. “We’re not trying to shut down open science,” he said. “We’re trying to make sure that the tools we’re building are safe and that the people who build them are accountable.”
For businesses evaluating AI tools, this is a critical distinction. Open-weight models can be incredibly valuable, they’re often more transparent, easier to audit, and more adaptable. But if those models are being developed in environments where safety and accountability are less prioritized, the risks may outweigh the benefits. Amodei’s comments suggest that businesses should look beyond the “open” label and consider the provenance, methodology, and governance behind any AI model they adopt.
This isn’t just about choosing between open and closed models, it’s about understanding the geopolitical context in which those models are being developed. The rise of Chinese AI labs, particularly those focused on distillation and model compression, is a trend that’s hard to ignore. And while Amodei doesn’t see a direct business risk in open-weight models from China, he’s not dismissing the broader implications.
One of the most interesting aspects of Amodei’s position is his willingness to acknowledge the value of open-weight models, even those from China, while still raising red flags. He’s not advocating for a blanket ban, nor is he dismissing the potential for innovation. Instead, he’s calling for a more nuanced approach, one that recognizes the benefits of openness while also being vigilant about the methods and contexts in which those models are being developed.
This is a conversation that’s going to continue to evolve. As AI becomes more integrated into business operations, the choices we make about which models to use, and from whom, will have real consequences. Amodei’s clarification is a reminder that the debate isn’t just about technical openness, but about how we want to govern AI development, and who gets to set the rules.
As first reported by TechCrunch. For more on how AI is reshaping the future of work, check out our post on The Quiet Engine Behind AI’s Next Leap: Materials Science.