Anthropic Says It Does Not Support Ban on Models With Open Weights
Anthropic CEO Dario Amodei said the company has never supported a ban on models with open weights, in a company blog post. The post was published on July 27, 2026, after reports that some US officials were considering limits on Chinese models with open weights.
Amodei wrote that models with open weights can benefit businesses, developers, and researchers when they do not have dangerous capabilities. He said a ban on their use by US companies would not address his main national security concerns, including the possibility that authoritarian governments build more powerful models or use AI for military and surveillance purposes.
He also said models with open weights can present higher misuse risks than closed models because their safeguards can be removed and their use is difficult to monitor after release. He argued that banning US business use would not stop bad actors, who are unlikely to be legitimate US companies.
Amodei said Anthropic supports three measures instead: restricting powerful chips and chipmaking equipment from reaching China, targeting industrial scale distillation operations, and requiring safety testing for all sufficiently capable models before release, whether they are open or closed.
We hope you enjoyed this article.
Consider subscribing to one of our newsletters like AI Policy Brief or Daily AI Brief.
Also, consider following us on social media:
More from: Regulation
More from: AI Safety
Subscribe to AI Policy Brief
Weekly report on AI regulations, safety standards, government policies, and compliance requirements worldwide.
Whitepaper
AI and the Law: Discussion Paper
This discussion paper explores the intersection of artificial intelligence and legal frameworks, addressing potential legal challenges posed by AI's autonomy, adaptiveness, and opacity. It examines issues such as liability gaps, causation, and the potential for granting AI systems legal personality, aiming to foster further discussion on AI's impact on law reform.
Read more