Anthropic CEO: We Don't Want Open-Weights AI Models Banned

Anthropic CEO Dario Amodei has publicly clarified his company's position on open-weights AI models, stating unequivocally that Anthropic has never advocated for banning them—a stance he felt compelled to make explicit amid recent controversy over Chinese AI development.

The statement comes as U.S. officials reportedly consider restricting American companies from using Chinese open-weights models, prompting tech companies to sign a letter defending open-weights development. Some observers had accused Anthropic of supporting such bans to protect its competitive position.

Why it matters

The debate over open-weights models sits at the intersection of AI safety, national security, and commercial competition. Amodei's position carries weight because Anthropic has been among the most vocal companies on AI safety issues, and his clarification distinguishes between legitimate security concerns and protectionist measures that wouldn't address actual threats.

Two nightmare scenarios

Amodei outlined two primary concerns that shape his thinking on AI policy, both detailed in his essay "The Adolescence of Technology" published six months ago.

First, he worries that authoritarian governments—particularly China's Communist Party—could develop AI models more powerful than those built by the U.S. and leverage them for military superiority or mass surveillance. He emphasized that this threat exists regardless of whether models are released with open weights, noting that the most dangerous scenario involves models trained in secret and deployed exclusively by military or intelligence agencies.

Second, he's concerned about powerful AI models being misused for cyberattacks or biological weapons development. While open-weights models do present higher risk in this category because guardrails are difficult to apply and releases cannot be withdrawn, Amodei argued that banning their use by U.S. businesses wouldn't address the problem since bad actors aren't likely to be legitimate companies anyway.

Three policy proposals

Instead of banning open-weights models, Amodei supports three specific measures:

Chip export controls: The U.S. should block sales of powerful chips and chipmaking equipment to China while cracking down on smuggling operations. Because China has limited domestic chip production capacity, this directly prevents it from training more powerful models than the U.S.

Stop industrial-scale distillation: Distillation allows China to build better models than its chip supply would normally enable, bringing its AI capabilities within months of the U.S. frontier. Amodei called for policy interventions targeting these operations, which are often backed by the Chinese state, while noting that a blanket ban on open-weights models is neither appropriate nor something Anthropic has requested.

Mandatory safety testing: All sufficiently capable models, whether open or closed, should undergo testing for cyber, biological, and alignment risks before release. Amodei noted this approach has gained traction in both the Trump administration and recent industry proposals, and suggested that even China might cooperate on preventing AI-enabled biological weapons since it's in their interest too.

Amodei acknowledged agreement with much of the industry letter defending open-weights models, including their benefits for access and competition. However, he disputed claims that open-weights models necessarily make it easier to develop safeguards or that broad access helps defenders more than attackers, pointing to potential attacker-defender asymmetries in biological threats.

The position statement was first published by Anthropic and reported by The Verge.