- Dario Amodei denies advocating for a ban on open-weight AI models, stating that Anthropic has never supported such measures.
- Amodei expressed concerns that authoritarian governments, particularly China, could develop AI systems that surpass those in the U.S. for military superiority or domestic repression.
- He emphasized that open-weights models without dangerous capabilities are a public good that provide value at minimal cost.
- Instead of a ban, Amodei proposed measures such as restricting chip sales to China and mandatory safety testing for AI models.
- Amodei highlighted that once open-weights are released, they cannot be withdrawn, citing a UK AI Security Institute report.
- He mentioned that distillation operations, which allow cheaper model training, pose a significant threat, particularly referencing a claim against Alibaba for a distillation attack.
Dario Amodei, CEO of Anthropic, has publicly denied advocating for a ban on open-weight AI models, stating, "We have not and are not advocating for a ban on open-weights models as a category." He emphasized that open-weight models without dangerous capabilities are a public good that provide value to businesses and researchers.135
Amodei expressed concerns about authoritarian governments, particularly the Chinese Communist Party, potentially developing AI systems that could surpass those in the U.S. He warned that such advancements could lead to permanent military superiority or be used for domestic repression. He stated, "The first centres on authoritarian governments, which he said could build AI systems more powerful than those developed in the US and use them either to secure lasting military advantage or to intensify domestic repression."2
Instead of a ban, Amodei proposed three measures: restricting the sale of advanced chips to China, cracking down on large-scale distillation operations, and introducing mandatory safety testing for all capable models. He noted that once open-weights are released, they cannot be withdrawn, citing a UK AI Security Institute report. Amodei also supports creating a model safety testing organization if global cooperation, including from China, can be achieved.46
On distillation, Anthropic has previously alleged that Chinese firm Alibaba conducted the largest known distillation attack against its systems, a claim raised in a letter to a US Senate committee last month.
“Amodei emphasized that open-weight models without dangerous capabilities are beneficial, providing value at minimal cost. He outlined specific measures to address concerns about authoritarian governments, including restricting chip access and implementing mandatory safety testing for AI models.”
TechCrunch
t2ONLINE
