Banning Chinese models won't help US win AI race: Zuckerberg
What's the story
Meta CEO Mark Zuckerberg has entered a major policy debate in the artificial intelligence (AI) space. He believes that banning Chinese models will not necessarily give the US an edge in the global AI race. His comments come as Washington considers new rules for powerful AI systems and industry leaders clash over open models, security risks, and China's technological ambitions.
Strategic focus
Focus on domestic competitiveness, not foreign restrictions: Zuckerberg
In an interview with Financial Times, Zuckerberg said that blocking advanced Chinese models from entering the US market wouldn't do much to help America in the global AI race.
He urged policymakers to focus on making domestic AI companies more competitive instead of putting restrictions on foreign competitors.
This comes as tech firms are divided over how governments should regulate increasingly powerful AI systems, especially with China's growing strength.
Innovation boost
Meta's stance on open-weight AI models
Zuckerberg reiterated Meta's long-standing support for open-weight AI models.
He believes that wider access to AI technology fosters innovation and enables developers, researchers, and businesses to create customized systems without relying solely on a few commercial providers.
However, he warned against the concentration of AI development in a few frontier labs.
While independent model reviews or safety assessments could benefit the industry, those leading the race shouldn't dominate these processes due to "regulatory capture."
Policy divide
NVIDIA CEO's letter sparks debate on AI model access
The debate has intensified after NVIDIA CEO Jensen Huang shared an open letter signed by several AI companies including Meta, Microsoft, and Hugging Face.
The letter called on governments not to impose what the signatories called "premature restrictions" on open-weight AI models.
Although it didn't mention China directly, the conversation has increasingly focused on fears that Chinese developers may be leveraging US intellectual property through methods like model distillation.
Clarification
Differentiating between open AI and frontier tech protection
Anthropic CEO Dario Amodei has clarified that his company doesn't want governments to ban open-weight models altogether.
Instead, he believes policymakers should differentiate between open AI and the separate issue of protecting frontier technology from theft.
He emphasized that openly available models with lower capabilities are an important public resource for research, software development, and broader innovation.
Safeguard measures
Calls for stronger safeguards against model theft
Amodei has called for stronger safeguards against model theft and continued export controls on advanced AI chips used to train the most capable systems.
The policy divide goes beyond open models as the Trump administration is likely to introduce a voluntary framework under which AI developers could submit their models for testing before public deployment.
Other industry leaders have also proposed new oversight mechanisms, with Google DeepMind chief Demis Hassabis advocating an industry-led organization that could establish common safety standards.
Cyber defense
Cyber defense concerns and Meta's proprietary shift
Zuckerberg warned that limiting access to frontier AI could unintentionally weaken cyber defense.
He cited a recent case of an OpenAI system allegedly going rogue and hacking a start-up, saying organizations without access to the most advanced proprietary models often turn to open-source alternatives for identifying vulnerabilities and developing security fixes.
Despite being one of the strongest proponents of open AI, Meta has also started testing proprietary offerings with its recently launched closed model Muse Spark.