Anthropic Says It Does Not Support Banning Open-Weight AI Models

Anthropic has clarified its position on open-weight AI models, with CEO Dario Amodei saying the company has never supported a ban. However, he continues to warn that powerful open models could create major security and geopolitical risks.

Anthropic raises concerns about China

Amodei warned that authoritarian governments, including China, could use advanced AI models to strengthen military capabilities, develop more powerful systems, and expand surveillance.

He also expressed concern that China could eventually build AI models that outperform those developed in the United States.

Anthropic has recently accused Alibaba of using 25,000 fake accounts to extract capabilities from Claude. Alibaba is also reportedly preparing to restrict Claude Code use across its workplace.

Open models could enable misuse

Amodei said bad actors could use powerful AI models to conduct cyberattacks and other harmful activities.

Open-weight models may present additional risks because developers cannot easily monitor how people use them. Once released, a company also cannot withdraw the model, update its safeguards, or prevent modified versions from spreading.

Amodei rejected claims that open models automatically make AI safer or consistently provide defenders with a greater advantage than attackers.

Amodei proposes four safety measures

The Anthropic CEO proposed several policies for managing advanced AI risks, including restricting exports of advanced chips and chipmaking equipment to China and increasing enforcement against chip smuggling.

He also called for banning industrial-scale AI model distillation operations and requiring mandatory safety testing for both powerful open and closed models.

These measures would target access to computing resources and require companies to evaluate advanced models before deployment or release.

Technology companies oppose early restrictions

NVIDIA, Meta, Microsoft, and other technology companies recently warned US policymakers against introducing premature restrictions on open-weight AI.

The companies argued that open models support economic growth, encourage innovation, and help researchers develop stronger AI safety tools. Anthropic did not sign the industry letter.

Amodei acknowledged that open-weight models can support research and economic growth. However, he argued that policymakers should not ignore their potential use in cyberattacks, military systems, surveillance, or the development of more capable AI.

In other Anthropic news, the company recently launched Claude Opus 5.