Anthropic CEO Voices Concerns Over Chinese Open-Weight AI Models Amid Silicon Valley Debate

A split has emerged in Silicon Valley over Chinese open-weight AI models. More than 50 companies, including Nvidia and Microsoft, have urged the White House not to ban these models, citing their strong performance and accessibility. However, Anthropic did not join this group. CEO Dario Amodei has clarified his company's position, stating that Anthropic is not against open-weight models but remains concerned about their potential misuse, especially by China.

Key Highlights

• Over 50 tech companies urged the White House not to ban Chinese open-weight AI models.
• Anthropic CEO Dario Amodei warns of risks if China gains access to powerful AI models.
• Amodei suggests targeted US restrictions on chip exports and industrial-scale AI distillation.
• Debate grows as Chinese AI models rival US systems in performance and cost.

Anthropic's Position on Open-Weight Models

In a recent blog post, Dario Amodei emphasized that Anthropic "never advocated for a ban on open-weights models." He acknowledged the benefits of open-weight models, noting they are cost-effective and valuable for businesses, developers, and researchers. Open-weight models like Kimi K3 and GLM 5.2 are cheaper to run than closed models such as Claude or GPT-5.6, and can be operated locally on personal devices.

Despite these advantages, Amodei expressed concerns about two major risks. He warned that authoritarian governments, particularly China, could exploit powerful AI models. Amodei stated, "The Chinese Communist Party (CCP) is clearly the most capable threat to build AI models that are more powerful than those built by the US." He fears that China could use advanced AI to achieve military superiority or suppress its citizens.

Risks and Proposed Restrictions

Amodei also highlighted the broader risks of open-weight models. As AI becomes more advanced, the potential for misuse increases, including cyberattacks and biological threats. He argued that open-weight models are harder to control because it is difficult to apply guardrails or monitor their use. Once released, model weights cannot be withdrawn, increasing the risk of misuse.

Anthropic has faced scrutiny over cybersecurity, particularly with the launch of Claude Mythos. The US government banned the Mythos and Fable models due to security concerns, though Anthropic later re-released them with added safeguards.

Instead of a blanket ban, Amodei suggests targeted restrictions. He recommends the US limit the sale of powerful chips and chipmaking equipment to China and address smuggling and workarounds. Despite US export controls, Chinese firms have continued to advance, with DeepSeek’s CEO noting that restrictions have pushed Chinese labs to collaborate with companies like Huawei to improve AI chips.

Amodei also calls for a crackdown on industrial-scale distillation, a process where smaller models are trained on the outputs of larger ones. He claims this allows Chinese labs to quickly approach US capabilities, though not surpass them. Additionally, he advocates for safety testing of all AI models to assess cyber, biological, and alignment risks before release.

Industry Reactions and Ongoing Debate

The debate has divided the tech industry. OpenAI, Nvidia, and Microsoft support open access to AI models, arguing it makes advanced AI more accessible and adaptable. Anthropic’s absence from the open-letter has drawn criticism, with some accusing the company of protecting its closed-weight business model.

Chinese startups like DeepSeek and Moonshot AI have released models that challenge US dominance, offering lower costs and competitive performance. This has intensified the policy debate in Washington, as US and Chinese firms compete for leadership in AI development.