Anthropic CEO Denies Ever Calling for Open Model Ban
Amodei's statement positions Anthropic against both a full open-model ban and the Microsoft-backed industry letter, carving a middle path that supports open weights while pushing for export controls and mandatory safety tests.
Key Facts
- Anthropic CEO Dario Amodei on July 27 denied that Anthropic has ever advocated banning open-weight AI models.
- The Little Tech Association, a group of about 200 companies including Proton and Y Combinator, has sent a letter to the Trump administration urging it not to regulate Chinese open-weight models.
- Anthropic declined to endorse a Microsoft statement supporting American-led open model development, which was backed by NVIDIA, OpenAI, Google, and AMD.
- Amodei said open-weight models without dangerous capabilities are public goods that cost nothing beyond computing resources to run.
Reporting from 2 sources: ASCII.jp, GIGAZINE.
Anthropic CEO Dario Amodei on July 27 denied that his company has ever advocated banning open-weight AI models, pushing back against regulatory debate that has included proposals to restrict foreign open models. In a statement released the same day, Amodei said open-weight models without dangerous capabilities are public goods that researchers, developers, and companies can use as long as they have computing resources. He framed his concerns around two national security risks: authoritarian states, particularly China, building high-performance AI to strengthen military and surveillance systems, and high-performance AI being misused for cyberattacks or biological threats. Amodei acknowledged that open-weight models are harder to guardrail after release, but argued that banning their use by U.S. companies would not stop malicious actors and would only restrict competition. He proposed three priority policies: strengthening export controls on high-performance AI chips and manufacturing equipment to China, cracking down on large-scale distillation, and mandating pre-release safety evaluations for capable models regardless of whether they are open or closed. He also responded to an industry letter supporting open models, agreeing it expands access and competition but cautioning against claims that open models favor defenders.
Amodei's statement comes as the Trump administration reportedly considers restrictions on foreign open-source AI models, and as the Little Tech Association, a group of about 200 companies including Proton and Y Combinator, has sent a letter urging the administration not to regulate Chinese open-weight models. Amodei said protective bans would not resolve his most serious national security concerns, describing two nightmare scenarios. The first is authoritarian governments, with the Chinese Communist Party as the most prominent threat, building AI models stronger than those made in the U.S. and using them for permanent military advantage or severe repression of their own people. He said the most dangerous models might be those secretly trained and given only to a national security ministry. The second is powerful AI being misused for cyberattacks or bioweapons, with open-weight models posing potentially higher risk because they cannot be revoked or monitored after release. However, he argued that banning U.S. companies from using these models would not address malicious actors, who are unlikely to be legitimate U.S. firms. On the Microsoft-endorsed statement supporting American-led open model development, Anthropic said it agrees with much of it but cannot accept the claim that open-weight models necessarily make safety measures easier or that broad access favors defenders over attackers, citing concerns about biological asymmetries where attackers could weaponize pandemic-level viruses faster than defenses can be built.
Synthesized by Yomimono from the 2 cited sources below, including Japanese-language reporting where cited, then editorially reviewed before publishing.