US-China AI race puts open models in spotlight; Anthropic CEO calls for AI rules instead of bans

Anthropic CEO Dario Amodei has opened up about U.S. restrictions on open-weights AI models and clarified that Anthropic does not have a ban. However, he expresses concerns about countries like China that could develop AI systems more powerful than those created in the US.

In a company blog post, Amodei highlighted that AI systems controlled by authoritarian governments could potentially be used for military advantages such as improving weapons systems, cyber operations, intelligence gathering, or strategic planning or domestic surveillance and repression.

Must read: US claims China's Moonshot AI used Anthropic's Fable 5 to build Kimi K3 AI model

“Anthropic has never advocated for a ban on open-weights models,” Amodei said. “Open-weights models that don’t have dangerous capabilities are a public good,” he added.

Anthropic CEO reveals national security concerns

In the blog post, Amodei said that open-weights models without dangerous capabilities could be beneficial in terms of providing substantial value to developers, researchers, and businesses at minimal cost. “Protectionist bans would not address my most serious national security concerns,” he said.

The executive also talked about two major national security risks associated with advanced AI, including the potential misuse of powerful models by authoritarian governments:

Authoritarian Advantage: Amodei says that his biggest concern is not around open or closed AI. He highlighted that the biggest worry is that countries with authoritarian governments could create AI systems that are more advanced than those developed in the US.

Must read: Anthropic signs supply deal with Samsung, SK hynix amid custom AI chip race

He highlighted the Chinese Communist Party (CCP) as the most capable potential threat, and such governments could use highly advanced AI for cyberattacks, military purposes, and use AI for mass surveillance, monitoring citizens, controlling information, or suppressing opposition.

Misuse & Alignment Risks: His second concern is around the misuse of advanced AI models. He said that “AI models may be misused to carry out cyberattacks or biological attacks, and may have serious alignment problems.” He says open-weight AI models may carry higher risks than closed models and that they cannot easily be taken back or controlled.

Amodei proposes policy actions

Amodei says that rather than banning open-weight models, he proposed three specific policy actions, which include restricting access to advanced chips and chipmaking equipment, limiting foreign authoritarian entities from using high-capability U.S. models, and risk assessment on all frontier models before public release.

However, he said that open-weight models bring benefits like more innovation and competition, but they should not automatically be considered safer.